7 OSCILLATORS In our study of RF transceivers in Chapter 5, we noted the extensive use of oscillators in both the transmit and the receive paths. Interestingly, in most systems one input of every mixer is driven by a periodic signal, hence the need for oscillators. This chapter deals with the design of oscillators for RF applications with emphasis on monolithic implementation. Beginning with some general issues, we consider basic oscillator topologies and study phase noise in oscillators. Next, we describe bipolar and CMOS oscillators and present methods of gen- crating quadrature and single-sideband signals. 7.1 GENERAL CONSIDERATIONS 206 An oscillator generates a periodic output. As such, the circuit must entail a self-sustaining mechanism that allows its own noise to grow and eventually become a periodic signal Most RF oscillators can be viewed as feedback circuits, Consider the simple linear feedback system depicted in Fig. 7.1(a), with the overall transfer function Ys) Hs) =. 71 XG) 1-A® om) A self-sustaining mechanism arises at the frequency sp if H (so) = +1, and the oscillation amplitude remains constant if so is purely imaginary, ic, H(s) = joo) = +1. Thus, for steady oscillation, two conditions must be simultaneously ‘met at wp: (1) the loop gain, |H (jap) |, must be equal to unity, and (2) the total phase shift around the loop, 4H (je»), must be equal to zero (or 180° if the de feedback is negative).Sec.7.1 General Considerations 207 ‘ x6)-@ Vis) X(s)> vis) + Frequency Selective Network (a) (b) Figare 71. (4) Feedback nator yt, () shin of equeneyseloeve aetwork oo Called Barkhausen’s criteria, the above conditions imply that any feed- back system can oscillate if its loop gain and phase shift are chosen properly. Ring oscillators and phase shift oscillators are examples. In most RF oscillators, however, a frequency-selective network, ¢g., an LC tank, is included in the loop so as to stabilize the frequency. This is illustrated conceptually in Fig. 7.1(b). although such a network can be part of H(s) as well. We deal with the issue of frequency stability in Section 7.4. The frequency-selective network is also called a “resonator, We should also mention that in some cases Barkausen’s criteria are nec- essary but not sufficient [2]. As a simple example, if the phase shift around the loop is equal to 360° at zero frequency and the loop gain is sufficient, the circuit latches up rather than oscillate. ‘The above view of oscillators is called the “two-port” model in microwave theory because the feedback loop is closed around a two-port network, eg, H(s) in Fig. 7.1(a). By contrast, the “one-port model” treats the oscillator as two one-port networks connected to each other [Fig. 7.2(a)]. To understand this model, suppose the resonator is a simple tank, shown in Fig, 7.2(b) along with its parasitic resistances. Recall from Chapter 2 that, for a narrow band of frequencies, the circuit can be converted to the parallel combination depicted in Fig. 7.2(c). The tank by itself does not oscillate indefinitely because some of the stored energy is dissipated in R, in every cycle. The idea in the one- port model is that an active network generates an impedance equal to — Rp so that the equivalent parallel resistance seen by the intrinsic, lossless resonator is infinite. In essence, the energy lost in R, is replenished by the active circuit in every cycle, allowing steady oscillation. ‘The two models of Figs. 7.1(a) and 7.2(a) are equivalent in many cases, with exceptions including inherently one-port active circuits such as tunnel diodes [3]. Oscillators utilized in RF applications often fall in the feedback category, but, where applicable, the one-port model can give additional insight into their operation.208 Chap. 7 Oscillators Ay Ry t e p= ox Bt Ay=-Re (a) (b) (©) Figure 72 (a) Oue port view of ori 2) LC rsonato() eivalnt circuit of (b). ‘The nominal frequency of oscillation is usually determined by the prop- erties of the circuit, e.g,, the resonance frequency of the LC tank in Fig. 7.2(c), But, how about the amplitude? The self-sustaining effect allows the circuit's noise to grow initially, but another mechanism is necessary to limit the growth at some point. For example, as the amplitude increases, the amplifier saturates, dropping the loop gain to a low value at the peaks of the waveform [1]. From another point of view, to ensure oscillation start-up, the small-signal loop gain must be somewhat greater than one, but to achieve a stable amplitude, the “average” loop gain must return to unity. Another mechanism used to define the oscillation amplitude is automatic level control (ALC). Illustrated in Fig. 7.3, this technique measures the ampli- tude by means of a peak detector (or a rectifier), compares the result with a reference, and adjusts the gain of the oscillator with negative feedback. Thus, at the start-up, the amplitude is small and the gain is high, and in the steady state, the amplitude is approximately equal to Veer. The use of ALC allows the oscillator to remain “linear” if Var is chosen properly, providing a sinu- soidal output with low distortion. However, the complexity of the circuit and the possibility of additional noise contributed by ALC have prevented wide usage of this technique in RF applications. Figure 7.3 Automatic level control.c.7.2 Basic LC Oscillator Topologies 209 What is the desired output waveform of an RF oscillator? Recall from the analysis of mixers in Chapter 6 that the noise contributed by the switching pair is minimized if the LO waveform exhibits abrupt transitions. Moreover, effects such as direct feedthrough are suppressed if the LO signal has a 50% duty cycle. Thus, the ideal waveform in this case is a square wave. In practice. however, circuit bandwidth limitations, particularly in LC oscillators, result in a waveform close toa sinusoid. Consequently, as shown in Fig. 7.4, the waveform is generated with a large amplitude to approximate a square wave in the vicinity of the zero crossings. Figure 7.4 Approximation of a square wave with a large sinusoid. In applications where the outputs of two or more oscillators are mixed to produce new frequencies, the desired waveform may be a low-distortion sinusoid. We return to this issue in Section 7.9. The need for a 50% duty cycle in most oscillators calls for differential topologies, an important point to remember in the study of various oscillator configurations. 7.2 BASIC LC OSCILLATOR TOPOLOGIES Most discrete RF oscillators incorporate only one active device, This is for two reasons: (1) to minimize the noise, and (2) to lower the cost, While the first reason still holds, the second is of little concern in IC technologies, The basic oscillators described in this section fall in the “one-transistor” category, serving as the first step toward understanding the design principles. Following the feedback model of Fig. 7.1(a), we postulate that a one- transistor LC oscillator may include an LC tank at the collector of a bipolar transistor, with the feedback signal applied to the base or the emitter (Fig. 7.5). (The same idea applies to FET implementations as well.) Let us assume that, as, shown in Fig. 7.6(a), the tank consists of an inductor and a capacitor in parallel Since at resonance, the impedance of the tank is real, the phase difference between its current and voltage is zero. Thus, to achieve a total phase equal to zero, the feedback signal must return to the emitter of the transistor. This is the basic idea in a number of oscillator topologies.210 Chap. 7 Oscillators @ oo) Figure 7.8. Feedback from (a) collector to base, (b) collector to emitter. ‘The connection of the tank to the emitter entails an important issue: the resistive loading seen at the emitter terminal, 1/gm. If the collector voltage is directly applied to the emitter, this resistance drastically reduces the Q of the tank, dropping the loop gain to below unity and preventing oscillation. For this reason, the emitter impedance must be transformed to a higher value before it appears in parallel with the tank [Fig. 7.6(b)]. (a) © Figure 7.6 (a) Direct feedback from collector to emitter, (b) insertion of an impedance transformer, (c) use of an explicit transformer. A simple approach to transforming the emitter impedance is to incorpo- rate a transformer in the tank, Fig. 7.6(c). For a lossless step-down transformer with a turn ratio 7, the parallel resistance seen by the tank is equal to n7/ 2m. Passive impedance transformation can also be accomplished through the use of tanks with capacitive or inductive dividers (Chapter 2). Illustrated in Figs. 7.7(a) and (b), the resulting circuits are called the Colpitts and Hartley os- cillators, respectively. From the derivationsin Chapter 2, we note that the equiv- alent parallel resistance in the tank is approximately equal to (1+ Ci/C2)*/&m in Fig, 7.7(a) and (1 + L2/L1)?/gm in Fig. 7.7(b), enhancing the equivalent Q by roughly the same factor.Basic LC Oscillator Topologies 211 @ ) Figure 7.7 (a) Colpitts and (b) Hartley oscillators, Let us now discuss the choice of the inductor and capacitor values and the properties of Q1. The resonance frequency is @, = (\/LeyCeq) |, where Leg and Coq are the equivalent inductance and capacitance in the parallel tanks of Figs. 7.7(a) and (b). From the point of view of achieving large voltage swings, it is desirable to maximize the value of the inductor. To understand why, first note that the Q of inductors is typically much lower than that of capacitors, making it possible to write the equivalent parallel resistance of the tank as Rp = (Leqw;)?/ Rs, where Rs models the loss of the inductor as a series resistance, In typical inductors, Leg and Rs scale proportionally; that is, if Leq increases by a factor m, then so does Rp. Since the impedance of the tank at resonance is equal to Rp, the voltage swing for a given bias current also increases by the same factor. If the inductor is off-chip, increasing its value also minimizes the effect of bond wire inductance, an important issue at frequencies above 1 GHz Maximizing the value of the inductor, however, entails two trade-offs. First, the self-resonance frequency of the inductor decreases, approaching the oscillation frequency of interest. Second, the tank capacitance becomes limited by device parasitics, making it difficult to vary the frequency by adding a variable capacitor to the oscillator. We return to these issues in Section 7.4. Transistor Q, is the primary source of noise in the oscillators of Fig. 7.7 and hence plays an important role. The base resistance thermal noise and the collector shot noise can be mimimized by increasing the size and decreasing the bias current of the transistor, respectively. However, the first remedy raises the parasitic capacitances, and the second lowers the voltage swing. Thus, a compromise is usually necessary. The bias current is typically chosen to yield maximum swing at node X without saturating Q;. The Colpitts oscillator incorporates only one inductor and is thus more frequently used than the Hartley circuit. Both oscillators, however, suffer from two drawbacks. First, for the emitter impedance of the transistor to have negli-212 Chap. 7 Oscillators gible effect on the Q of the tank, the ratio of the capacitors or inductors tends to be a large number. Second, these oscillators provide only a single-ended output. One-Port View | While we have arrived at Colpitts and Hartley topolo- gies through a feedback model of oscillators, a one-port representation is also possible here, providing more insight into the behavior of these circuits. Con- sider the circuit shown in Fig. 7.8(a), where My is biased in saturation. Note that none of the nodes is grounded, What is the small-signal impedance Zin? As shown in Fig. 7.8(b), we can write the sum of 7p; and the current through Cras Ei got (2 py Gia = if 12 —2* s = Ix, : as® Ge) re G2) where the gate-drain capacitance has been neglected. Thus, Vx _ 8m ti a= + : 73 Ix Ge * Gs ed For § = je, this impedance contains a negative real part equal to —gin/ (CiC2e), indicating that the circuit can oscillate if an inductor is placed be- tween the drain and the gate [Fig. 7.8(c)]. Zz 5 &® ket My ey VA * ar or fa) (b) 1 & = 1 e a by Figure 78 (a) One-port active circuit, (b) calculation of input impedance, (¢) connection of an inductor in parallel with the port, Since the above derivation assumed no ground connection in the cireu we can obtain three different topologies by grounding one of the three termi- nals. This is illustrated in Fig. 7.9. 7.3 VOLTAGE-CONTROLLED OSCILLATORS: The frequency of most RF oscillators must be adjustable. For example, the front-end downconversion and upconversion functions studied in Chapter 5 must select one of many channels because a given transceiver is assigned dif- ferent carrier frequencies at different times, Thus, the LO frequency in each case must vary in well-defined steps (e.g., 30 kHz in IS-54). If the output fre- quency of an oscillator can be varied by a voltage, then the circuit is called a “voltage-controlled oscillator” (VCO). While current-controlied oscillatorsSee. 7.3 Voltage-Controlled Oscillators 213 4 i Te. Ho, 4 . “bin Lg, “T To ta] @ (b) © Figure 7.9 Oscillator topologies obtained by grounding one terminal of the cireuit in Fig. 7.8(¢). Biasing not shown.) are also feasible, they are not widely used in RF systems because of difficulties in varying the value of high- Q storage elements by means of a current. How is the frequency of an oscillator varied? In LC implementations, (part of) the tank capacitance can be provided by a reverse-biased diode (“var- actor”) so that the de voltage across the junction controls the resonance fre- quency. Illustrated in Fig, 7.10 are two tank configurations incorporating a varactor diode. , ba Veont v, feont Figure 7.10 Varactor diodes added to a tank, In our studies of phase noise and phase-locked synthesizers, we will need a mathematical model of VCOs. We define an ideal VCO as a circuit that generates a periodic output whose frequency is a linear function of a control voltage, Veont? out = Ore + Kyco Veom, where wpe is the “free-running” frequency and Kyco is the “gain” of the VCO (specified in radis/V). The existence of #7 in the above equation simply indicates that, for the practical range of Veont, @out May not approach zero, In other words, Voom creates a change around core. Since phase is the integral of frequency with respect to time, the output of a sinusoidal VCO can be expressed as y(t) = Acos (rat + Kveo f You . (7A)214 Chap.7 Oscillators For example, if Von: is a constant voltage Vo, then y(t) = Acos[(wpr + KycoVo)t + $0], where do represents the initial value of the phase. ‘Thus, as expected, if Von: is constant, the frequency is simply shifted by Kyo Veom- Equation (7.4) implies that a VCO is in fact a frequency modulator, and the results obtained in Section 3.2 apply here as well. In particular, for sinu- soidal modulation, veo (t) = Vm COS @yqt, We have veo K y(t) = Acos (ore + Ven sino) r (75) On indicating that the VCO has a tendency to reject high-frequency components that appear at its control input. Also, if |KycoVm/@m| & 1 rad, then the narrowband FM approximation (Chapter 3) applies, and the output spectrum consists of a main component at apg and two sidebands at wp R + mr 7.4 PHASE NOISE As other analog circuits, oscillators are susceptible to noise. Noise injected into an oscillator by its constituent devices or by external means may influence both the frequency and the amplitude of the output signal. In most cases, the disturbance in the amplitude is negligible or unimportant, and only the random deviation of the frequency is considered. The latter can also be viewed as random variation in the period or deviation of the zero crossing points irom their ideal position along the time axis. For a nominally periodic sinusoidal signal, we can write x(t) = A cosfwet + Gn (t)], where },(f) is a small random excess phase representing variations in the period. The function ¢,(t) is called “phase noise.” Note that for |@,(t)| K 1 rad, we have x(t) © A cos wet — Ag, (¢) sin w,4; that is, the spectrum of y(t) is translated to +t. In RF applications, phase noise is usually characterized in the frequency domain. For an ideal sinusoidal oscillator operating at a, the spectrum as- sumes the shape of an impulse, whereas for an actual oscillator, the spectrum exhibits “skirts” around the carrier frequency (Fig. 7.11). To quantify phase Ideal Oscillator Actual Oscillator ® ® Figure 7.11 Output spectrum of ideal and actual oscillators,Sec.7.4 Phase Noise 215 noise, we consider a unit bandwidth at an offset Aw with respect to «,, calculate the noise power in this bandwidth, and divide the result by the carrier (average) power. For example, if the carrier power is —2 dBm and the noise power measured in a 1-kHz bandwidth at an offset of 1 MHz is equal to —70 dBm then the phase noise is specified as —70 dBm +2 dBm —30 B= —98 dBe/Hiz, where dBc means “in dB with respect to carrier.” 7.4.1 Effect of Phase Noise in RF Communications To understand the importance of phase noise in RF systems, consider a generic transceiver as depicted in Fig. 7.12, where a local oscillator provides the carrier signal for both the receive and uransmit paths, If the LO output contains phase noise, bath downconverted and upconverted signals are corrupted. This is illustrated in Fig. 7.13. Frequency Synthesizer Figure 7.12 Generic transceiver front end Referring to the ideal case depicted in Fig, 7.13(a), we note that the signal of interest is convolved with an impulse and thus translated to a lower (and a higher) frequency with no change in its shape. In reality, however, the wanted signal may be accompanied by a large interferer in an adjacent channel, and the local oscillator exhibits finite phase noise (Fig. 7.13(b)}. When the two signals are mixed with the LO output, the downconverted band consists of two overlapping spectra, with the wanted signal suffering from significant noise due to the tail of the interferer. This effect is called “reciprocal mixing.” Shown in Fig. 7.13(c), the effect of phase noise on the transmit path is slightly different. Suppose a noiseless receiver is to detect a weak signal at «> while a powerful, nearby transmitter generates a signal at @ with substantial phase noise. Then, the wanted signal is corrupted by the phase noise tail of the transmitter. ‘The important point here is that the difference between «; and a can be as small asa few tens of kilohertz while each of these frequencies is around 900 MHz or 1.9 GHz. Therefore, the output spectrum of the LO must be extremely sharp. For example, in IS-S4, the phase noise power per unit bandwidth must be about —115 dBefHz.at 60 kHz offset.216 © Chap.7 Oscillators Lo Wanted (# Output Signal | ® Wanted Signal. a —— — Downconveried ” Oo Downconverted™ oO Signal Signals @ ©) Nearby Transmitter ae © Figure 7.13 (a) Downconversion by an ideal oscillator, (b) reciprocal mixing, (c) effect of phase noise in transmitters. To arrive at some representative phase noise specifications, let us consider the example depicted in Fig. 7.14, where the wanted channel is 30 kHz wide and 60 dB below an unwanted channel 60 kHz away. How low should the phase noise of the unwanted channel at 60 kHz offset be so that the SNR in the desired channel exceeds 15 dB? The total noise power introduced by the interferer in the desired channel is equal to fu Pro = fo Sa Adds (76) ft where S,(/) denotes the phase noise profile of the unwanted channel and fi. and fy are the lower and upper ends of the desired channel, respectively. For simplicity, we assume S,(f) is constant in this bandwidth and equal to So, obtaining Py,to: = So(fu — fi). Thus, Pug SNR = Se — 9" a7See.7.4 Phase Noise 217 tote t 60 kHz Figure 7.14 Calculation of signal degradation due to phase noise. yielding 1010g(So/ Pig) = —15 dB —10log( fy — fx). Since the interferer power, Pig, is 60 dB higher than Py, we have 10 log(So/ Pig) = —15 dB —10log( fi; — fi.) — 60 dB. It follows that the phase noise must not exceed —115 dBc/Hz at 60 kHz offset. Note that while the number —115 dBc/Hz seems quite small, it rises to approximately —70 dBe when integrated in a 30- kHz band, If the phase noise varies substantially from fi, to fy, then it must be specified at several frequency offsets and the integration in (7.6) must be carried out accordingly. Phase noise of a local oscillator also corrupts the information carried in the phase of the carrier. For example, the downconversion of a QPSK waveform by a mixer that is driven by a noisy LO results in the constellation shown in Fig, 7.15 [4]. Thus, the bit error rate in the detector may increase significantly. Figure 7.18 Effect of phase noise an QPSK signal constellation 74.2 Q of an Oscillator The phase noise of LC oscillators usually depends on their Q. Intuitively, we expect that the higher the Q of the tank, the sharper the resonance and the lower the phase noise skirts. It is instructive to examine three definitions of Q within the context of oscillators. Basic physics defines the Q as 27 x (energy stored/energy dissipated per cycle). For an LC tank, Q is an indication of how much of the energy is lost as it is transferred from the capacitor to the inductor and vice versa.218 Chap.7 Oscillators Resonant circuits usually exhibit a bandpass transfer function. The Q can also be defined as the “sharpness” of the magnitude of the frequency response. More specifically, as shown in Fig. 7.16, Q is defined as the resonance frequency divided by the two-sided —3-dB bandwidth. These two definitions yield the same value for a simple LC tank. |AGoy| Figure 7.16 A definition of Another definition of Q that proves especially useful in oscillators is illustrated in Fig. 7.17. Here, the circuit is considered as a feedback system and the phase of the open-loop transfer function, @(w), is examined at resonance. The Q is then defined as do Zl Q (7.8) | Called the “open-loop Q” herein, this definition has an interesting interpreta- tion if we recall that for steady oscillation, the total phase shift around the loop must be zero. Now suppose the oscillator frequency slightly deviates from a Then, if the phase slope is large, a significant change in the phase shift arises, thus violating the condition of oscillation and forcing the frequency to return to 9. In other words, the open-loop Q is a measure of how much the closed-loop system opposes variations in the frequency of oscillation. H(jo) o=/H(jo) Od o Figure 7.17 Definition of @ based on open-loop phase slope Let us apply (7.8) to the Colpitts oscillator of Fig. 7.18(a) to find the Q in terms of the circuit component values. We open the loop at the drain of ‘My, inject a current into the tank, and calculate the resulting current in Mj. Following the analysis of Section 7.2and assuming the Q of the tankis limited bySec.7.4 Phase Noise 219 the transformed value of 1/21, we utilize the equivalent circuit of Fig. 7.18(b) to write Vi i By Wr, (79) Ta ACA? , where Coq © CiC2/(Ci + C2) and Ry © (14+C1/C2)?/8m. (Ci + C2) > nm; then the voltage applied to the source of M; is approximately equal to Vyx[C2/(Ci + C2)], and the resulting drain current Zou = —mVx[C2/(Ci + C2)]. It follows that the open-loop transfer function is “65 = mene (1s Cas &). (7.10) ‘The term inside the parentheses gives a phase shift of et = = — nt _ 22 ___, (7.1) 2 RG — LC?) and hence, at o = wy = 1/\/ECaq db aa lente = —2CeqRp = (7.12) Thus, Q = RpCeqo = Rp/(Leo), the same value as obtained in the above two definitions. Note that we have neglected the phase shift in the transistor itself. T 1.3 c. I =A, g HO? ) Figure 7.18 (a) Open-loop Colpitts oscillator, (b) equivalent circuit of the drain network. In this example, we have assumed that Rp arises only from the input resistance of the transistor. In reality, the inductor and the capacitor exhibit their own loss, further lowering the Q, The overall Q is sometimes called the “loaded Q” to distinguish it from the Q of the tank itself, emphasizing the resistive component introduced by the transistor and any buffer stage following the oscillator.220 Chap. 7 Oscillators 7.4.3 Phase Noise Mechanisms Phase noise in oscillators has been studied extensively [5, 6, 7]. Effects such as nonlinearity and periodic variation of the circuit parameters make the analysis, of phase noise quite difficult. Nonetheless, for simple LC oscillators linear approximations have been used judiciously, yielding reasonable errors in the prediction of phase noise [6, 7]. More rigorous analysis of oscillators is still a topic of active research, Oscillator phase noise is generated primarily through two mechanisms, distinguished by the path into which the noise is injected. Recall that a VCO includes both a (feedback) oscillation signal path and a frequency control path. Illustrated in Fig. 7.19, the noise, x(t), appearing in these paths gives rise to distinctly different effects. We examine each case separately. Yeont Vas noise CD) vo 2@-+[ro] y(t) His o Vout @ oy Figure 7.19 Phase noise in (a) signal path, (b) control path, Noise in Signal Path Consider the circuit of Fig. 7.19(a), where the noise is injected into the oscillation signal path. How is y(#) affected by x(t)? Representing the open-loop circuit by a linear transfer function H(s), we can write YG) _ HG) ¥@ “iow (7.13) Inthe vicinity of the frequency of oscillation, «2 = w-+Aw, we can approximate H (j) with the first two terms in its Taylor expansion: dH H(jw) © H(joo) + do. (7.14) do Since H (ja) = +1 and typically |Awd H /do| < 1, (7.13) reduces to ¥ yeu + Ao) ~ —Fe (7.15)See. 74 Phase Noise 221 Output Spectrum Noise Power Spectral Density i x aA o Wy o Figure 7.20 Noise shaping in oscillators implying that a noise component at» = a + Aq is multiplied by —(AedH /de)~! when it appears at the output of the oscillator, In other words, as shown in Fig. 7.20, the noise spectrum is shaped by [25] ? 1 = ——_,,. (7.16) (awy| 4 do Fu (wy + Ae)] Let us now consider the term |d H /dw|?. Expressing H (w) in polar form, H(@) = |Hlexp(jd), we have dH _(dlHi , . db : ied (= + SIAN Z, ) expe). (7.17) It follows that d\H\ dof is [ | de + Te Hy. (7.18) Equation (7.18) can be used to arrive at a general definition of Q [25]. but we limit our study here to simple LC oscillators. We make two observations. First, in an LC oscillator the term |d|H|/de|? is much less than |d@ /de|? in the vicinity of the resonance frequency. Second, the value of |H| is close to unity for steady oscillations. Thus, |dH/des|* ~ |d¢ /de|*, allowing (7.16) to be written as a 7.19) XO) ~ Face fa to 8) 2 do) From (7.8), we have 2 1 poy? = ig (=) (7.20)Chap.7 Oscillators Known as “Leeson’s equation” [6], this expression reveals the dependence of the output noise upon the Q of the tank, the center frequency, and the offset frequency. Note that (7.20) includes the effect of noise on both the amplitude and the phase of the carrier, The phase noise is typically half the value given by (7.20). It is important to note that the overall phase noise power with respect to the carrier depends on two other parameters as well: the noise generated by the devices, that is, the magnitude of X (j@) and the amplitude of oscillation, Vs. In the oscillators of Fig. 7.7, for example, the noise contributed by Q) arises from the base resistance and the collector current Ic), while the amplitude is given by the collector current swing and the parallel equivalent resistance Rp. Since the noise current increases in proportion to J/7ci, whereas Vs is proportional to Jc: Rp, typical designs maximize Jc, without saturating the transistor. From the above study, we can deduce an interesting property of LC oscil lators. Asthe Q of the tank increases, three aspects of the performance improve simultaneously: (1) the noise shaping function, (7.20), becomes sharper; (2) the power dissipation decreases; and (3) the noise injected by active devices de- creases. To understand the last two, note that the equivalent parallel resistance of the tank Rp, & Q, and hence the voltage swing Vs x Ici - Q. ‘Thus, a higher Q allows a lower Icy to achieve the same voltage swing, thereby saving power. Furthermore, as /c; decreases, the noise current injected by the tran- sistor decreases. For these reasons, a great deal of effort has been expended on increasing the Q of inductors [10, 11]. Our analysis of the noise in the signal path has thus far assumed a linear feedback system. As mentioned in Section 7.1, oscillators usually experience amplitude limiting and hence nonlinearity, thus “folding” the noise compo- nents, Illustrated in Fig. 7.21, this effect arises when odd-order nonlinearity in the amplitude leads to intermodulation between an injected noise component at cw, and the carrier, creating another component at 2 — @, [25]. In order to represent the effect of noise folding due to nonlinearity, Leeson’s equation can be multiplied by a factor [8, 9] is > x07 ~), (721) do where A is the actual small-signal loop gain. This relation indicates that the loop gain must be chosen as close to unity as possible while ensuring reliable start-up. A typical value is between 2 and 3. Noise in Control Path In the mechanism studied above, the noise in- jected into the signal path simply mixes with the carrier. By contrast, noise injected into the control path affects the frequency by changing the physical properties of the oscillator. For example, if a varactor is used to tune the VCO, noise on the de voltage applied across the diode varies the tank capacitanceSec.7.4 Phase Noise Amplifier and Resonator Figure 721 Noise folding in an oscillator. and hence the resonance frequency. Viewed as analog frequency modulation, this effect translates low-frequency noise components in the control path to the region around the carrier [Fig. 7.22(a)] ‘To quantify the FM noise mechanism, as depicted in Fig. 7.22(b), we represent the noise per unit bandwidth as a sinusoid with the same average power: V»COS@q/. Denoting the gain of the VCO by Kyco and using the narrowband FM approximation, we have AoVmnK vco Vout (t) & Ap COS wot + 20m [£08 (co + @m)t — cos( — Gm Jt}. (7.22) ‘Thus, the noise power at w+ oy with respect to the carrier power is equal to (Kyco/@m)*Vq/4. In practice, Kyco is proportional to the carrier frequency because for a given control voltage range, the tuning range must be a constant, percentage of the center frequency so as to compensate for process and tem- perature variations. Note that the effect of this type of noise becomes more prominent as @», decreases, making 1/f noise in the control path particularly detrimental. X%y { { font Veo =O 0 Ong 1 OO @o-@p Doty (a) A. 0 On pease 0 On ) Figure 722 (a) Modulation of VCO frequency by noise on control line, (b) approximation of noise with a sinusoid,224 Chap.7 Oscillators We should also mention that the FM noise mechanism arises from not only the explicit control path of a VCO, but any source that can vary the frequency of oscillation. An interesting example occurs in oscillators employing discrete inductors whose value may vary with mechanical vibrations. As another ex- ample, noise in the collector current of Q; in Fig. 7.7 varies the base-emitter diffusion capacitance, thus modulating the resonance frequency. ‘The mechanisms formulated in this section predict that the phase noise decreases indefinitely as Aw increases. In reality, however, the noise reaches a relatively fiat floor because the loop does not shape the noise of the devices at high frequency offsets. 7.4.4 Noise-Power Trade-off As other analog circuits, oscillators exhibit a trade-off between power dissipa- tion and noise. Intuitively, we note that if the output voltages of N identical oscillators are added in phase (Fig, 7.23), then the total carrier power is multi- plied by N, whereas the noise power increases by N (assuming noise sources of different oscillators are uncorrelated). Thus, the phase noise (relative to the cartier) decreases by a factor N at the cost of a proportional increase in power dissipation. ®o Figure 723 Addition of the output voltages of V oscillators. Our study of phase noise indicates that the following parameters must be taken into account: the center frequency, @p, the power dissipation, P, and the offset frequency, Aw. Thus, the phase noise power of different oscillators must be normalized to (w)/Aw)*/P for a fair comparison. 7.4.5 Effect of Frequency Division and Multiplication on Phase Noise Since frequency and phase are related by a linear operator, division of frequency bya factor N isidentical to division of phase by the same factor. For anominallySee.7.4 Phase Noise 225 periodic sinusoid, x(t) = A cos[to.¢+4n(t)], where ¢n (F) represents the phase noise, a frequency divider simply divides the total phase by N, ©, bnlt) Acos| Sr + 2], 7. os [¢ y / (7.23) where the phase noise contributed by the divider is neglected. This indicates that the magnitude of phase noise at a given offset frequency is divided by N. and from narrowband FM approximation the phase noise power is divided by N?, X1/N Interestingly, the above trend is observed even if an explicit frequency divider is not used. For example, in the LC oscillators of Fig. 7.7, if the total tank inductance and capacitance are increased by a factor N, the frequency of oscillation decreases by the same factor while the Q, the sources of noise, and the amplitude remain relatively constant. Thus, from (7.20), the phase noise power at a given offset drops by N?. A similar analysis shows that frequency multiplication raises the phase noise magnitude by the same factor. 7.4.6 Oscillator Pulling and Pushing In our study of phase noise, we have tacitly assumed that the magnitude of the noise injected into the signal path is much less than that of the carrier, thereby arriving at a noise shaping function for oscillators. An interesting phenomenon is observed if the injected component is close to the carrier frequency and has, a comparable magnitude (Fig. 7.24). As the magnitude of the noise increases, the carrier frequency may shift toward the noise frequency @, and eventually “lock” to that frequency. Called “injection pulling,” this effect is described in (12, 13]. Injected Signal gp © > Po ao 209-On, 209-On Oy Po 3 The © 209-Op Op 2@o-On, Figure 7.24 Injection pulling of an oscillator as the noise amplitude increases.226 Chap.7 Oscillators In a transceiver environment, various sources can introduce oscillator pulling. For example, as described in Chapter 5, the power amplifier output may couple to the local oscillator(s). Another example of injection pulling arises in the receive path when the desired signal is accompanied by a large interferer (Fig. 7.25). If the interferer frequency is close to the LO frequency, 1g through the mixer may pull 9 toward qg,. Thus, the VCO must be followed by a buffer stage with high reverse isolation. As explained in Chapter 6, however, the noise of such a stage may increase the noise figure of the mixer. Interferer Low Noise pases ‘Amplifier Signal 0 Bint “@ Figure 7.25 Injection pulling due to large interferer, Another type of oscillator pulling is illustrated in the GFSK modulator of Fig. 7.26, a topology used in DECT transmitters [14]. Here, the VCO is first placed in a feedback loop (Chapter 8) so as to stabilize its output frequency. Subsequently, the control voltage of the VCO is switched to the baseband signal, allowing the VCO to function as a frequency modulator. However, the ‘open-loop operation makes the VCO frequency quite sensitive to variations in the load impedance (“load pulling”). In particular, as the power amplifier turns on and off periodically to save energy, the VCO center frequency changes considerably, demanding a high-isolation stage between the VCO and the PA [14]. Baseband Data ° Figure 7.26 Load pulling due to variation of impedance. RF oscillators in general exhibit a poor supply rejection. In Fig. 7.10, for example, if the supply voltage varies, so do the reverse voltage across the var- actor and the frequency of oscillation. In RF design, this is called “supply push- ing.” An important case of supply pushing occurs in portable transceivers whenSec.7.5 Bipolar and CMOS LC Oscillators 227 the power amplifier turns on and off. Owing to the finite output impedance of the battery, the supply voltage may vary by several hundred millivolts. 75 BIPOLAR AND CMOS LC OSCILLATORS The stringent phase noise requirements in RF systems generally necessitate os- cillator topologies incorporating passive resonators. The resonators are built as very narrowband SAW or crystal filters, transmission lines, or simply LC tanks, the phase noise typically increasing in that order. For many applications, LC tanks having a @ of greater than approximately 20 prove adequate. Re- call from Section 7.4.3, however, that the relative phase noise also depends on sources of noise and the oscillation amplitude. To minimize the former, low- noise oscillators usually employ only a few active devices in the signal path. As with other analog circuits, the design of VCOs becomes more diffi- cult as the supply voltage scales down. At lower supplies, the voltage swings are smaller, thereby yielding a higher relative phase noise, Furthermore, the reverse bias voltage range of the varactor is narrower, thus limiting the tuning range. Note that since at high frequencies, parasiticcapacitances are significant, the varactor can change only a fraction of the tank capacitance. ‘The limited tuning range in LC oscillators has been viewed as both a merit and adrawback! On the one hand, since K yo isrelatively small, the frequency is less sensitive to noise on the control line. On the other hand, additional means of adjusting the frequency—at least during assembly—are required to ensure operation in the band of interest despite manufacturing variations. In many RF systems, this is accomplished by adding a small mechanically trimmable capacitor in parallel with the varactor, although the tuning range must still be wide enough to cover variations with temperature. However, the present trend toward higher levels of integration demands that such adjustments be eliminated by achieving a sufficient tuning range. 75.1 Negative-G,,, Oscillators As explained in Section 7.2, the signal fed back from the collector (or drain) of a transistor to its emitter (or source) must pass through an impedance trans- former so as to avoid loading the tank excessively. In the Colpitts and Hartley oscillators of Fig. 7.7, the impedance is transformed by means of passive net- works. Now consider the oscillator shown in Fig. 7.27(a) where an active buffer, By is interposed between the collector and the emitter, presenting a relatively high impedance to the tank. As depicted in Fig. 7.27(b), the buffer can be implemented as an emitter (or source) follower. Note that the base of Q1 is connected to Vec to have the same de voltage as the base of Q2. (The two transistors are assumed to be identical.) The circuit can also incorporate two inductors to operate differentially (Fig. 7.27(c)]. However, if the inductors are off-chip, bond wire parasitics make it difficult to achieve asmall phase mismatch between the two tanks.228 © Chap.7 Oscillators Yeo @ () © Figure 7.27 (a) Addition of active buffer in feedback loop of a Colpitts oscillator, (b) implementation of buffer with an emitter follower, (c) differential implementation of (b). ‘The feedback oscillator of Fig. 7.27(c) can be viewed as a one-port imple- mentation as well. Calculating the impedance seen at the collector of Q; and Q) as in Fig. 7.28, we note that positive feedback yields Rin = —2/g,,. Thus, if [Rin] is less than or equal to the equivalent parallel resistance of the tank, the circuit oscillates. This topology is called a “negative-G,, oscillator.” Figure 7.28 Circuit to calculate the input impedance of cross-coupled pair.Sec. 7.5 Bipolarand CMOS LC Oscillators 229 In addition to contributing noise, Q1 and Q» in Fig. 7.27(c) saturate heav- ily if the peak swing at nodes X and ¥ exceeds approximately 400 mV. To re- solve this issue, a capacitive divider can be inserted in the feedback path (Fig. 7.29), allowing greater swings and higher common-mode level at nodes X and ¥ than at the bases of Q; and Q> Figure 7.29 Capacitive division in feedback path of an oscillator. ‘The negative-G,» oscillator can also be implemented in MOS technology Shown in Fig. 7.30 are three examples [15, 16, 17]. Since large swings are quite benign here, no capacitive divider is required in the feedback paths. The circuit of Fig. 7.30(c) achieves differential operation with only one inductor but at the cost of higher noise due to J; and I. Tuning Issues _In order to control the oscillation frequency in the above circuits, varactor diodes must be added to the tanks. Fig. 7.31(a) shows an ex- ample where the junction capacitance of Dj and D2 is controlled by Veont. In.@ typical CMOS technology, D; and D; are implemented asshown in Fig.7.31(b). * To understand the tuning range limitations of this circuit, let us assume the rep- resentative values L) = Lz = 10 nH, (W/L); = (W/L)2 = 100 xm/0.6 wm, Ipp = 2 mA, and Vpp = 2.7 V, and a center frequency of 1.8 GHz, The di- mensions of M;-Mz and the value of Jp» are chosen so as to provide sufficient loop gain for Q © 4. How wide is the tuning range of this circuit? We make three observations. First, for integrated inductors in a three-metal layer process, each of Ly and Ly exhibits a distributed parasitic capacitance of approximately 0.5 pF, which can be lumped into two 0.25 capacitors connected from each terminal of the inductor to ground. Second, the total capacitance contributed by My and My and a typical output buffer to each of the nodes X and ¥ is approximately equal to 0.25 pF. Third, diodes D; and D; are accompanied by roughly 0.1 pF of capacitance to the p-substrate.230 © Chap.7 Oscillators aI “HI ey b) Figure 7.30 CMOS oscillators, (a) output referred to Vop, (b) output referred to ground, (c) differential operation with one inductor. Inductor Parasitics p-substrate (a) o) Figure 7.31 (a) Addition of varactors to control the frequency, (b) imple- mentation of varactors in CMOS technology.Sec.7.5 Bipolar and CMOS LC Oscillators. 231 From the above observations, we can calculate the total fixed capacitance in each tank to be roughly equal to 0.6 pF. For f = 1.8 GHz and L = 10 nH, this leaves only 0.18 pF for the variable component of the varactor capacitance. By what factor can this capacitance change? Recall that the drain voltage swings are large so as to minimize the relative phase noise. Thus, the minimum. value of Vx and Vy is typically less than 0.5 V, indicating that Veon, must rem: below approximately 1.2 V to avoid forward biasing D; and D3. Furthermore, if the oscillator operates within a frequency synthesizer, the minimum value of Veout, determined by the previous stage, is no less than a few hundred mili- volts. Such limited range of Veon typically varies the diode capacitance, C;, by approximately 50%. Since in this example the junction capacitance of each varactor constitutes only 23% of the overall tank capacitance, a 50% change in C; translates into less than 10% adjustment in the oscillation frequency. It is interesting to note the trade-off between the tuning range and the relative phase noise. If the voltage swings at nodes X and ¥ in Fig. 7.31(a) are reduced, Voom can assume higher values, thereby varying the frequency by a greater amount. Also, if L; and L2 are decreased, the total tank capacitance and hence the contribution of the varactors can increase, widening the tuning range, but at the cost of lower voltage swings (or a higher bias current) 75.2 Interpolative Oscillators The underlying assumption in the VCOs studied thus far has been that the frequency can be varied by only adjusting the tank capacitance (or inductance). Interpolative oscillators provide an alternative tuning mechanism by employing more than one resonator. Shown in Fig, 7.32 is a feedback oscillatory system incorporating two transfer functions, H;(s) and H2(s), whose outputs are scaled by variable fac- tors ay and a2, respectively, and summed. The overall open-loop transfer Og Figure 732 _Interpolative oscillator.Chap. 7 Oscillators function is therefore equal to H(s) = a; Hy(s) + a2H2(s), which must equal +1 for the system to oscillate. In the extreme cases where a; = 0 or a2 = 0, the oscillation frequency, a, is determined by only H(s) or H2(s), and for intermediate values of @; and a2, can be “interpolated” between its lower and upper bounds. Of course, this occurs only if the roots of H(s) — move in a “well-behaved” manner from one extreme to the other. ‘To understand the principle of interpolation and its limitations, let us con- sider the circuit depicted in Fig. 7.33, a conceptual illustration of the oscillator described in [18]. In this circuit, Ry and R; denote the equivalent parallel resis- tance of each tank, For the system to oscillate, the open-loop transfer function must equal unity: GmRiLis Gyo RoLos Ee BE agg jee GBR "REGS + Ls + Ri | RLaGs + Ls + 4) If (7.24) has a unique set of conjugate roots on the imaginary axis or in the right half plane, then the circuit oscillates at a single frequency. Figure 733 Interpolative oscillator using two tanks. What happens if the resonance frequencies of the two tanks, @ and @, are far apart? Figure 7.34 plots the open-loop gain and phase of such an os- cillator [18], revealing wo frequencies at which the total phase is zero and the gain is greater than unity. As a consequence, the circuit may oscillate at more than one frequency. It can be shown that the maximum spacing between «> and q22 that avoids this effect is equal to («, + @2)/(2Q), where the two tanks are assumed to have equal Q’s[18]. Interestingly, interpolative oscillators, too, exhibit a trade-off between the phase noise and the tuning range.Sec. 7.6 Monolithic Inductors. 233 |AGod| 1 © [H(jo) ° © Figure 734 Open-loop gain and phase of an interpolative oscillator. An important concern in interpolative oscillators is the noise due to the additional active devices in the signal path, particularly in the adder. Nonethe- Jess, the combination of capacitor tuning and interpolation may provide a wide tuning range. 7.6 MONOLITHIC INDUCTORS Fabrication of inductors on the same substrate as the rest of an RF circuit can obviate the need for external connections, thus resolving issues such as electrical and magnetic coupling and pad and bond wire parasitics, Furthermore, the addition of inductors to the IC designers’ device library can lead to new circuit techniques that relax the trade-offs in the RF design hexagon. ‘The most important parameters of monolithic inductors are the Q, the self-resonance frequency, and the area, all of which strongly depend on the layout and the properties of the IC technology. Inductors are usually imple- mented as spiral structures (Fig. 7.35). A circular shape exhibits less metal resistance for a given value of inductance and metal wire width. The circle may be approximated by a hexagon or an octagon [19] Figure 7.38 Rectangular spiral inductor.234 Chap.7 Oscillators The value of spiral inductors can be obtained by means of various accurate [20] or approximate [21, 22] methods. An empirical formula that has less than 10% error for inductors in the range of 5 to 50 nH is given in [17] and can be reduced to the following form for a square shape, ag? L © 13 x 10-7" ____, Are W75(W + GP25 (7.28) where A», is the metal area (the shaded area in Fig. 7.35), Atot is the total inductor area (~ S? in Fig. 7.35), W is the line width, and G is the line spacing. Allunits are metric. Prediction of the Q of inductors is much more difficult. As shown in Fig, 7.36, the loss in a monolithic inductor results from three mechanisms: metal wire resistance, capacitive coupling to the substrate, and magnetic coupling to the substrate [23]. Determined by the substrate resistivity and the size of the inductor, the last two effects limitthe Q athigh frequencies. Note that magnetic coupling becomes more significant in CMOS technologies with heavily doped substrates because the effect of substrate resistance appears in parallel with the inductor. Ve Ly Z “mene Pie substrate susetate “HONE 8 @ @) Fsub substrate © Figure 7.36 Loss mechanisms in monolithic inductors, (a) resistive, (b) capacitive, (c) magnetic. ‘The effect of capacitive coupling to the substrate can be reduced by plac- ing under the inductor a grounded “shield” (Fig. 7.37). Since a continuous plateSec.7.7 Resonatorless VCO's 235 Figure 7.37 Broken shield plate to reduce capacitive loss mechanism, would increase the magnetic coupling, the plate must be broken regularly in the direction perpendicular to the current flow. In a spiral inductor, the small internal turns contribute little inductance while suffering from all of the loss mechansims. Thus, removal of the first four or five turns may increase the Q slightly. ‘The self-resonance frequency, fs, sets an upper limit on the value and dimensions of the inductor for a given operation frequency. The distributed structure of spirals makes it difficult to estimate fs, usually requiring elec- tromagnetic field simulations or actual measurements. Note that even if the operation frequency is below fs, each inductor exhibits significant capaci- tance to the substrate. Despite extensive recent work, the Q of inductors in standard CMOS technologies has been limited to less than 5. Research on monolithic inductors nevertheless continues. 1.7 RESONATORLESS VCO’S ‘The Barkhausen criteria for oscillation can be met without resonators in a circuit as well. If the open-loop circuit exhibits sufficient gain at the zero-phase frequency, oscillation occurs. Also, if the open-loop input/output characteristic entails hysterisis, the circuit may oscillate even if the small-signal phase shift does not appear to be sufficient. An example of the first type are ring oscillators. As shown in Fig. 7.38, if a cascade of M gain stages with an odd number of inversions is placed in a feedback loop, the circuit oscillates with a period equal to 2MTy, where Ty is the delay of each stage with a fanout of one. The oscillation can be viewed as occurring at the frequency for which the total phase shift is zero and the loop gain is unity. Various implementations of the gain stages utilized in ring oscillators along with methods of tuning the frequency are described in [24]. (Odd Number of Inversions aes ieee al Figure 7.38 Ring oscillator.236 Chap. 7 Oscillators Figure 7.39shows an example of oscillators incorporating hysterisis. Called a “relaxation oscillator,” the circuit can oscillate even if C, is the only storage element, simply because the positive feedback around M; and M provides rapid switching at the end of each cycle, breaking the feedback loop for most of the period. At high frequencies, however, the hysterisis diminishes and oscillation occurs only if the poles at nodes X and Y contribute significant phase shift [25]. It is interesting to note that the snrail-signal impedance seen looking into the X-¥ port of the cross-coupled pair is approximately equal to —2/gm — 1/(Cis), that is, a negative resistance in series with a negative capacitance. In high-frequency operation, the first component cancels the loss in Ry and Rp, and the second resonates with the capacitance at X and Y. Yoo Ry Re x ¥ My, Mo Cc. Iss * @lss Figure 7.39 Relaxation oscillator. Resonatorless oscillators have not been popular in RF design. ‘This is because they not only exhibit an open-loop Q close to unity but contain many noisy active and passive devices in the signal path. For example, in a three-stage differential ring oscillator, the open-loop Q is approximately equal to 1.3, and nine transistors (including the tail current sources) and six load resistors add noise to the carrier [25]. 7.8 QUADRATURE SIGNAL GENERATION Inourstudy of transceiversin Chapter 5, we noted the frequent need for quadra- ture LO signals in both the receive path and the transmit path. In this section, we describe methods of generating signals with a phase difference of 90” 7.8.1 RC-CR Network A simple quadrature technique is to shift the signal by +45° using an RC-CR network (Fig. 7.40). Asexplained in Section 5.2.3, the phase difference between Voutt and Vou is 90° for all frequencies, but the output amplitudes are equal only at = 1/(RC). Thus, ifthe absolute value of RC varies with temperatureSec.7.8 Quadrature Signal Generation 237 Figure 740 Quadrature network using RC-CR circuit, and process, so does the frequency at which equal-amplitude quadrature sig- nals exist. In the LO path, where the information lies in only the zero-crossing points, the amplitudes can be equalized by means of “limiting” stages, for ex- ample, differential pairs, Amplitude limiting nevertheless becomes difficult in gigahertz circuits unless several stages are placed in cascade. Under these con- ditions, however, the phase (and gain) mismatch of the chains of limiters in the Tand Q paths becomes significant. In addition, limiting stages entail “AM-to-PM conversion.” In general, a nonlinear circuit with finite bandwidth exhibits a delay that depends on the input slew rate [26]. To understand this effect, consider the simple example depicted in Fig. 7.41, where the capacitors are added to represent the limited bandwidth, and other parasitics are neglected. We examine the circuit’s phase shift for small and large input amplitudes. Figure 741 Limiter with finite bandwidth For a small-amplitude sinusoid with frequency @ applied at the input, the output differential current is also close to a sinusoid, thus experiencing a phase shift equal to |9,| = tan~'(R,C,w). Now suppose the input is a sinusoid with a large amplitude such that M; and M; rapidly switch at each zero crossing of Vig. In this case, the differential output current is close to a square wave, resulting in a delay equal to RyC; In2 and hence a phase shift equal to |@| = RiCyw In 2.Chap. 7 Oscillators From this analysis, it follows that the phase shift varies from RyCyw In2 to tan™!(R)C,w) according to the slew rate of the input. This effect is particu- larly noticeable if the input has amplitude modulation, In the limiting process, the phase is disturbed by the amplitude variation, and we say AM-to-PM con- version occurs. The key point in the context of quadrature generation is that a difference in amplitudes translates to a phase imbalance if the limiting stages do not provide adequate bandwidth. In the circuit of Fig. 7.40, mismatches between resistors and capacitors deviate the output phase difference, @, from 90°. Assuming a relative resistor mismatch of @ and capacitor mismatch of 8, we can express @ in the vic of o = 1/(RC) as ¢ = — [tan“' RO + aC. + 6) — tan“! RCo] (7.26) _ my RCo + all + A) — RCo = 27" TH RCod +a + ARCO Gan ay 2 egg te & > ~ tant (7.28) we 22 OE =F 7 (7.29) For example, if « = 6 = 1%, then the phase imbalance is equal to 3.6/ (27) © 0.6" It is interesting to note that the phase and amplitude imbalances in Fig. 7.40 do not depend on the additional load capacitance seen at Voyx; and Vour2 10 ground. This is because such capacitances only impact the poles of the circuit, whereas the phase difference arises from the zero in the upper path. However, capacitive paths between the two outputs do introduce phase error, demanding careful layout. Simulations indicate that if this capacitance is 2% of C, then a phase imbalance of 1° appears, Mismatches between the load capacitance also contribute to phase error. Another issue in the quadrature circuit of Fig. 7.40 is the harmonic con- tent of vin. Suppose vin(t) = Ay coset + A, cosnot. To shift vin(t) by 90°, we replace ¢ with t — 7/4, where T = 2/w. Thus, vint — T/4) = A; cos(@t ~ 7/2) + Ay cos(ncot — nz/2), indicating that the nth harmonic must be shifted by 7/2. In the RC-CR network, on the other hand, the phase shift is equal to 90° for all frequencies, thereby resulting in a phase imba ance between Uoyrt and voy. Furthermore, the magnitude of the harmonics also experiences unequal gains through the two paths, introducing amplitude mismatches at the output.Sec. 7.8 Quadrature Signal Generation 239 ‘To quantify the effect of the second harmonic, we express the outputs as AL VRC 1 fo VIR Ca +1 AiR ARE cog [or = tan1(RCw) + =| VRC0 + 1 2 2A2RC@ + Co Vouti (t) cos[wr — tan7'(RCw)] cos(2wt — tan” !(2RCw)] (7.30) Vounlt) = cos [an = tan“! QRCo) + For a nominal value of RC = w7', we have AL nm), A 4 Voui(t) = —= cos (wt — —) + —cos(2ot — t 2) 7.32) WO=% (or - 2) cosa — tan (732) At mY) , 242 “in. veces] — he t+ =) + 2 cos (2or — tan-!2 Yourtt on (eo +3)+ 7 008 (208 aes) (7.33) Simulations predict that for the phase error to be less than 1°, we must have Az < 1.4%). If only the third harmonic is present, then it must be less than 1.2% of the fundamental for a phase error of 1°, If the LO harmonics are significant, the RC-CR network must be preceded by a lowpass filter. The value of C in Fig. 7.40 is usually chosen so as to minimize the atten- uation due to the load capacitance; for example, C © SCz. Large values of C, however, lower the input impedance of the circuit, which is approximately equal to 1/Cs. 78.2 Havens’ Technique Ilustrated in Fig. 7.42(a), Havens’ technique first splits the signal by approxi- mately 90°, generating V; and V3, and subsequently adds and subtracts these two phases, producing Vous and Voy2 [27]. As shown in Fig. 7.42(b), if V; and V> have equal amplitudes, the angle between Vou and Vou is equal to 90° This can also be seen by writing vi(t) = A cose, v(t) = A cos(wt +8), and hence 8 ult) + v(t) = 2A cos = cos (or + 5) (7.34) wits) — (1) = 2A sin 2 sin (0: + (7:35)240 © Chap.7 Oscillators Vouti oe Figure 742. (a) Havens quadrature circuit, (b) phasor dlagram of the operation ‘The limiting stages following the phase shift circuit equalize the amp) tudes of v; and v3. Moreover, since the adders’ outputs have different ampli- tudes if @ # 90°, they too are applied to limiters. In both cases, the issue of AM-to-PM conversion becomes important. If the signals sensed by the two adders in Fig, 7.42(a) exhibit amplitude mismatch, then the outputssutfer from phase imbalance. To quantify this effect, let us assume v2(t) = (A + €) cos(wt + 4), where € represents the mismatch, and redraw the phasor diagram of Fig. 7.42(b) as shown in Fig. 7.43(a). We note that the effect of € cos(wt + @) is to rotate Your counterclockwise by $1 and vou clockwise by ¢2. With the aid of the diagram in Fig. 7.43(b), we can write _ esin(6/2) tan bi = 54 cos(@/2) + €cos(0/D) (36) ee ee 737) 2A sin(6/2) + € sin(@/2)"See. 7.8 Quadrature Signal Generation 241 Voutt esin(0/2) ecos(8/2) 2A.cos(0/2) 2asin(g/2) e sin(0/2) Vourz ecos(0/2) @ oo) Figure 7.43 (a) Effect of amplitude mismatch in the Havens circuit, (b) simplified version of (a). Thus, for ¢ < A, we have € [sin(@/2) cos(/2) we £ [sim@s2) , c08(0/2) 3 athe xy [Sen * eon os € £ ae a 3 Asing er) Since @ is in the vicinity of 90°, (7.39) suggests that a 1% amplitude mismatch translates into 0.6° of phase error. ‘The above analyses have assumed sinysoidal signals. However, both the limiting operation following the 90° phase shift stage and the nonlinearity in the two adders create harmonics of v;(r) and v2(¢), Furthermore, the input signal itself may contain harmonics as well. We first consider the effect of the second harmonic. Suppose vj (t) and v2(¢) contain second harmonies with equal amplitudes, for example, acos2wt and acos(2wr + 29), respectively. Subtraction and addition of such components still yields two signals with a frequency of 2 that are 90° out of phase. However, recall from Section 7.8.1 that if Uouti and vou are to be in quadrature, then the nth harmonic must have a phase difference of nz /2. Thus, the existence of two second-harmonic components with 90° phase difference results in a phase imbalance between the two outputs. The effect of odd-order harmonics is less troublesome. Suppose, after limiting v(t) = A coswt + acos(2k + lot and v(t) = Acos(ot +) +242 Chap.7 Oscillators acos[(2k + 1jwt + (2k + 1)6), where k is an integer. Upon addition and subtraction, we obtain 1) = 2Acos! 2 You(t) = 2A cas 5 cos { ot + > (2k +16 [ Qk +1) OS cos + 2a cos 2k + Det + Selo 40) # = waa! 242 Your(t) = 2Asin > sin (wr + > 2k +1 + 2asin EU ein [e + Der + 4741) (Qk +1) “—e 2 For Vout and vous to have a phase difference of 2/2, we must have Vouz(t) = Vou lt — 7 /(2@)}. Calculating this term, x e 6 (2k +18 a) = Z =) + 2acos +" vou (t Fe) = 2Acos 5 sin (o + 4) + 2a cos 2 sin [oe + Dor + ADE - i]. (7.42) we note that the second terms in (7.41) and (7.42) differ in two respects: the amplitude and the polarity. Thus, odd-order harmonics introduce amplitude mismatch but no phase imbalance. Another source of amplitude mismatch in the Havens topology is the capacitive coupling between the two inputs of the adders, Illustrated in Fig, 7.44 is an example, where v; and v2 are applied to limiting stages and subsequently an adder consisting of two differential pairs. The gate-drain overlap capacitance of the transistors provides a path from the I channel to the O channel, in essence pulling the quadrature phases together. Similar to capacitive feedthrough from Vout 10 Yous? in Fig. 7.40, this effect results in significant phase imbalance at the inputs of the adders if the output impedance of the limiters is not sufficiently low. Figure 7.44 Capacitive path in an adder.Sec. 7.9 Single-Sideband Generation 243 78.3 Frequency Division A simple approach to generating quadrature periodic signals with frequency is to use a master-slave flipflop to divide a signal at 2a, by a factor 2 (Fig. 7.45). If Vin has a 50% duty cycle, then Vogt and Voui2 are 90° out of phase. Various implementations of divide-by-two circuits are described in Chapter 8. Figure 71.45 Frequency divider as a quadrature generator. The principal difficulty in this approach is that generation and division of the signal at 2a may consume substantial power or simply be impossible due to technology limitations. Another issue is the phase imbalance resulting from deviation of the input duty cycle from 50%. For example, if Vi contains asecond harmonic, then Vouxi and Voy exhibit a phase mismatch similar to that described for the RC-CR network of Fig. 7.40.-Mismatches in the signal paths through the latches also contribute phase error. 7.9 SINGLE-SIDEBAND GENERATION ‘Transceiver architectures often require the addition or subtraction of the out- put frequencies of two or more LOs. For example, the offset-LO technique described in Section 5.3.1 and employed in the GSM system of Fig. 5.49 mul- tiplies two signals and selects the sum or difference frequency by a bandpass filter. However, as shown in Fig. 7.46(a), if one of the frequencies is much less than the other, the filter must have a sharp cutoff characteristic so as to suppress. the unwanted sideband. An alternative technique of generating various LO frequencies is to em- ploy single-sideband architectures. Shown in Fig. 7.46(b), the idea is based on the identity cos af coset + sin ant sinant = cost F w2)t, requiring quadrature phases of both frequencies. This topology, however, is susceptible to two types of imperfections: mismatches between the upper and lower paths and nonlinearities in the mixers. The unwanted sideband resulting from phase and gain mismatches is explained in Section 5.3. Thus, we consider only the nonlinearities in the signal path. Recall from Chapter 6 that mixers are usually designed such that the LO port experiences abrupt switching; that is, the LO port is so nonlinear that the RF signal is multiplied by a rectangular waveform. Furthermore, the RF port244 Chap. 7 Oscillators Figure 7.46 (a) Simple mixing, (b) ideal SSB mixing. exhibits nonlinearity even with degenerative feedback. As a consequence, in Fig, 7.46(b) harmonics of @; and @ are generated in each port of the mixers, leading to various cross-products after multiplication. Since the unwanted sidebands in this circuit can be maintained below roughly —30 dB with respect to the desired sideband, the use of an SSB gener- ator greatly relaxes the cut-off characteristics of the bandpass filter required in Fig. 7.46(a). More sophisticated calibration and linearization techniques may even obviate the need for an external filter. REFERENCES [1] B.vanderPol, “The Nonlinear Theory of Electric Oscillations,” Proc. IRE, Vol. 22, pp. 1051-1086, September 1934, [2] N.M. Nguyenand R. G. Meyer, “Start-up and Frequency Stability in High- Frequency Oscillators,” JEEE Journal of Solid-State Circuits, Vol. 27, pp. 810-820, May 1992. [3] H.L. Kraus, C, W. Bostian, and F. H. Raab, Solid State Radio Engineering, New York: John Wiley, 1980, [4] B. Law and M. Groh, “Identifying RF-Related Impairments in Full- Service Digital Networks,” Microwave J., pp. 88-94, March 1996.Chap. 7 References 245 [5] E. Hafner, “The Effect of Noise in Oscillators,” Proc. IEEE, Vol. 54, pp. 179-198, February 1966. [6] D. B. Leeson, “A Simple Model of Feedback Oscillator Noise Spectrum,” Proc. IEEE, Vol. 54, pp. 329-330, February 1966. [7] W.P. Robins, Phase Noise in Signal Sources, London: Peter Peregrinus, 1982, [8] B. Jansen, K. Negus, and D. Lee, “Silicon Bipolar VCOs for 1.1 to 22 GHz with Fully-Integrated Tank and Tuning Circuits,” ISSCC Dig. of Tech. Papers, pp. 392-393, February 1997. [9] C. A.M. Boon, Design of High-Performance Negative Feedback Oscilla- tors, Ph.D. Thesis, Delft University, The Netherlands, 1989. [10] N.M. Nguyen and R. G. Meyer, “SiIC-Compatible Inductors and LC Pas- sive Filters.” JEEE Journal of Solid-State Circuits, Vol. 25. pp. 1028-1031, August 1990, [11] R. B. Merril et al., “Optimization of High Q Inductors for Multi-Level Metal CMOS,” Proc. IEDM, pp. 38.7.1-38.7.4, December 1995. [12] K. Kurokawa, “Injection Locking of Microwave Solid-State Oscillators,” Proc. IEEE, Vol. 61, pp. 1386-1410, October 1973. [13] R. Adler, “A Study of Locking Phenomena in Oscillators,” Proc. IRE, Vol. 34, pp. 351-357, June 1946, [14] R. Mohindra, “Isolator for DECT Open-Loop Modulator,” RF Design, pp. 30-42, January 1996. [15] B. Razavi, “A 1.8 GHz CMOS Voltage-Controlled Oscillator,” ISSCC Dig. Tech. Papers, pp. 388-389, February 1997, [16] B. Razavi, “A 900-MHz CMOS Direct-Conversion Receiver,” VLSI Cir- cuits Symp. Dig. of Tech. Papers, Sune 1997. [17] J. Craninckx and M. 8. J. Steyaert, “A 1.8 GHz CMOS Low Phase Noise Voltage-Controlled Oscillator with Prescaler,” [EEE Journal of Solid- State Circuits, Vol. 30, pp. 1474-1482, December 1995. [18] N.M. Nguyen and R. G. Meyer, “A 1.8-GHz Monoloithic LC Voltage- Controlled Oscillator” IEEE Journal of Solid-State Circuits, Vol. 27, pp. 444450, March 1992 [19] J. Craninckx, M. Steyaert, and H. Miyakawa, “A Fully Integrated Spiral LC CMOS VCO Set with Prescaler for GSM and DCS-1800 Systems,” Proc. CICC, pp. 403-406, May 1997. [20] H. M. Greenhouse, “Design of Planar Rectangular Microelectronic In- ductors” JEEE Trans. on Parts, Hybrids, and Packaging, Vol. 10, pp. 101-109, June 1974, [21] F. E. Terman, Radio Engineers’ Handbook, New York: McGraw-Hill, 1943,