Autocalibration of VCO frequency and loop gain is an essential process in PLL frequency synthesizers. In a wide tuning-range fractional-N PLL frequency synthesizer, high-speed and high-precision automatic calibration is especially important for shortening the lock time and improving the phase noise. This paper reviews the design issues of the PLL auto-calibration and discusses on the limitations of the previous techniques. A very simple and efficient auto-calibration method based on a high-speed frequency-to-digital converter (FDC) is proposed and verified through simulations. The proposed method is highly suited for a very wide-band ΔΣ fractional-N PLL.
1. Introduction
ΔΣ fractional-N RF frequency synthesizer is an essential building block in modern wireless communication systems. Achieving wide frequency tuning range, low phase noise, and fast locking time is challenging especially as the required tuning range becomes wider. If a single tuning curve is used to cover total tuning range, the VCO gain KVCO of an LC VCO needs to be extremely large as illustrated in Figure 1(a). As a result, the phase noise and spur performances will be unacceptably poor. This is why multiple subband tuning curves via a switched capacitor bank are usually needed to cover the wide tuning range as well as keep KVCO low, as shown in Figure 1(b) [1]. In this case, if the total capacitance of the switched capacitor bank varies in linear proportion to the cap bank code n, the KVCO and the sub-band spacing fstep will vary in a cubic power of the fVCO variation [2]. For example, KVCO and fstep vary eight times if fVCO varies two times. Due to such a wide variation of the VCO tuning characteristics, the PLL’s essential performance parameters such as loop gain, lock time, and phase noise vary much, which is usually undesirable. Therefore, a fast and accurate autocalibration is needed for a PLL frequency synthesizer.
Frequency tuning characteristics in a wide band VCO. (a) Single tuning curve. (b) Multiple tuning curves with KVCO(n) and fstep(n) variation.
Finding a sub-band tuning curve that is the closest to a target frequency is referred to as the VCO frequency calibration. There had been many methods reported in the literature [3–10]. However, they all showed severe speed-resolution limitation. Meanwhile, maintaining the loop gain constant over the entire tuning range is referred to as the loop gain calibration. It is required in order to have a constant loop bandwidth and low phase noise. Previous techniques of this also showed slow calibration time due to the closed loop operation [11, 12].
This paper reviews the previous calibration techniques and examines their limitations. Based on the considerations, a new autocalibration technique based on a high-speed FDC is presented, which is highly suited to a wide tuning range ΔΣ fractional-N synthesizer.
2. Previous Techniques and Design Considerations2.1. VCO Frequency Calibration
The most critical issue for the VCO frequency calibration is the accuracy and speed, that is, the calibration resolution and time. The calibration time should be fast enough because it will enlarge the total lock time, and the calibration resolution should be at least better than fstep/2 in order to be able to resolve two adjacent sub-band tuning curves. Note that fstep can be easily smaller than fREF in a fractional-N PLL. Thus, the calibration circuit should be able to provide sub-fREF calibration resolution while the calibration time is as fast as possible.
The tuning voltage monitoring technique as shown in Figure 2(a) was the oldest technique [4–6]. It monitors the VCO tuning voltage Vtune to find if it goes out of the predefined range between VHIGH and VLOW. If it goes out of the range, the cap bank code is automatically adjusted. Figure 2(b) illustrates the operation procedure. First, the calibration starts at point A on code n-1. Then, Vtune goes up toward VHIGH due to the high ftarget. When Vtune exceeds VHIGH, the calibration circuit changes the cap bank code from n-1 to n. Then, PLL settles into the lock state with fVCO approaching ftarget. If the calibration process starts from point B on code n+1, similar process is carried out to reach the cap bank code n. This method always maintains the closed-loop lock state during the calibration process. Because of the closed-loop operation, this method usually requires several hundreds of μsec for the calibration, which is the major drawback of this method. This method, however, can be useful when it is used as an auxiliary lock maintaining tool to deal with the unwanted VCO tuning characteristic variation during the normal closed-loop operation of PLL [6].
VCO frequency calibration by Vtune monitoring. (a) Architecture. (b) Operation principle [4–6].
Next the most popular VCO calibration technique is the “relative” frequency comparison method [7, 8]. The structure is shown in Figure 3(a). It compares the counted values of two clock signals fREF and fVFC during k·TREF counting period. As illustrated in Figure 3(b), fVFC(=fVCO/M) is counted during k·TREF. Then, the comparator simply compares the two values and generates Fast/Slow flag signal, and then the control logic changes cap bank code accordingly. During the calibration, PLL stays in the open-loop state and Vtune is fixed at VDD/2. The frequency resolution of this technique is given by M·fREF/k. As in the usual implementation, if M is the PLL’s total division ratio N, the frequency resolution will be fVCO/k. Thus, k must be higher than N in order to have sub-fREF resolution, and higher k leads to a longer calibration. Thus, the typical calibration time of this method usually reaches several tens of μsec or even hundreds of μsec, which is also a major drawback of this method.
Relative frequency comparison VCO frequency calibration. (a) Architecture. (b) Operation principle [7, 8].
Another very fast but more analog intensive method is the time-to-voltage converter- (TVC-) based “relative” period comparison technique [9, 10]. The schematic structure is shown in Figure 4(a). It compares the voltages VC1 and VC2 that are proportional to the periods of the two incoming clock signals fREF and fVCO/N (Figure 4(b)). This method is usually very fast and only requires a few μsec or even sub-μsec. But due to the analog circuitry involved in the comparison process, its accuracy can be easily affected by the circuit mismatches and PVT variations. In addition, it is not suitable to a ΔΣ fractional-N PLL. Figure 5 illustrates how the fDIV clock edges can vary when ΔΣ modulator is activated to provide a fractional division ratio N·f. Assuming an MASH-111 is used for the ΔΣ modulator, fDIV clock edge can be anywhere between (N-3)·TVCO and (N+4)·TVCO depending on the ΔΣ modulator output values, where TVCO is a VCO signal period in lock state. Therefore, if the TVC-based technique is applied to a ΔΣ fractional-N PLL, a multiple period integration of the TVC output is needed, which will not only make the calibration time slow but also degrade the calibration accuracy.
Relative period comparison VCO frequency calibration. (a) Architecture. (b) Operation principle [9, 10].
f
DIV clock period jittering at the lock state with a ΔΣ modulator activated in a ΔΣ fractional-N PLL.
Another issue in the VCO frequency calibration methods is about the code selection algorithm. Note that the conventional simple binary search process only gives an odd numbered code as a final result [7]. In Figure 6, code 9 would be selected as a final code by the conventional simple binary search process although even-numbered code 10 is the closest to the target frequency f1. In order to overcome this, an improved method in [10] compares the last two searched codes. But this algorithm will also fail if the closest code is in multiples of 4. This limitation is explained in Figure 6. This method is not a problem for the target frequency f2, for which the closest code 2 will be selected. But for the target frequency f3, the method in [10] will provide the less optimal code 3 rather than the optimal code 4. Therefore, a truly closest code selection algorithm must be able to store the frequency error for all searched codes during the calibration process and finally produce the closest code to the target frequency.
Final code selection of binary search process for various target frequencies in previous techniques. For f1, typical binary search process is used [7], and improved algorithm in [10] is applied for f2 and f3.
2.2. Loop Gain Calibration
A charge pump PLL generally uses the third-order passive loop filter as shown in Figure 7(a). It is known that the loop gain is proportional to ICP·KVCO·R2/N, the lock time is proportional to √(ICP·KVCO·R2/N), and the integrated phase noise is inversely proportional to ICP·KVCO·R2/N. Thus, the variation of each parameter will cause significant variation of the PLL closed-loop performances such as the loop gain, the lock time, and the phase noise. Figure 7(b) depicts the loop bandwidth variation due to the loop gain variation. The effect of ICP variation on the loop gain can be minimized by using the same-type on-chip resistor for the loop filter R2 and the ICP reference voltage generation [15]. The division ratio N is a digital value, so its impact on the loop gain can be precisely predicted and accurately compensated. Meanwhile, the variation of KVCO due to PVT variation cannot be accurately predicted. Therefore, the loop gain calibration circuit must be able to gauge the KVCO variation accurately, which should be equivalent to the loop gain variation.
(a) A third-order passive loop filter. (b) Closed-loop gain and bandwidth variation in PLL.
Figure 8(a) shows the previous loop gain compensation method [13]. Note that the term “compensation” is used here instead of the preferred “calibration.” The basic idea of this compensation is to make the charge pump gain follow a preset way such that it is inversely proportional to fVCO2. Figure 8(b) illustrates the operation principle of this compensation. To compensate KVCO/N variation that is proportional to fVCO2 in an LC tuned VCO, the charge pump gain is controlled to be inversely proportionally to fVCO2. Then, the loop gain that is proportional to ICP·KVCO/N will stay at a constant value, resulting in the constant loop bandwidth. But since this method cannot deal with any unexpected PVT variation of KVCO, its effect would be limited after the chip fabrication.
Loop gain compensation by preset pattern of charge pump gain ICP. (a) Architecture. (b) Operation principle [13].
The limitation of the previous method in [13] comes from the absence of the on-chip gauge of the loop gain. The on-chip measurement of the loop gain and the use of this value for loop gain calibration should be a very effective approach. It is known that PLL’s closed-loop time-domain step response is closely related to the PLL’s loop bandwidth. Previously, using the time-domain step-response to measure the loop gain and calibrate the charge pump gain was reported [11, 12]. Figure 9(a) shows the block diagram and Figure 9(b) shows the operating procedure. To calibrate loop gain, the two-step integrator integrates the VCO output signal in the locked state. After that, the step input Δf is triggered, and the integrator integrates the PLL’s step-response. In [12], only the step-response is integrated to reduce the calibration time. The drawback of this calibration method is a long calibration time, that is almost up to several tens of μsec due to the closed-loop operation. In addition, the accuracy might be affected by the KVCO variation during the transient state. Thus, we need more accurate and fast on-chip KVCO measurement tool for the loop gain calibration.
Loop gain calibration by measuring step-response of PLL (a) Architecture. (b) Operation principle [10, 11].
2.3. Linearization of VCO Coarse Tuning Characteristics
The variation range of KVCO and fstep are known to be proportional to the cubic power of the fVCO total tuning range as the following equation:
KVCO,maxKVCO,min=fstep,maxfstep,min=(fVCO,maxfVCO,min)3.
The large variation of KVCO and fstep degrades the speed and accuracy of the loop gain and VCO frequency calibration processes. Therefore, reducing the total variation of KVCO and fstep across the entire tuning range is desirable for improving the calibration result. Several studies were reported to reduce the KVCO and fstep variation in LC VCO: authors of [16, 17] only addressed the KVCO issue, and authors of [18] proposed a new cap bank structure to reduce the KVCO and fstep variations together, but it relied on arbitrary fractional scaling of the capacitance value over the sixteen tuning curves, which made it difficult to apply the method to an even wider tuning range and more tuning curves. Thus, we need more simple and systematic design method of the capacitor bank structure for reducing the KVCO and fstep variations across the total tuning range.
3. FDC-Based Autocalibration Technique
The limitations of the previous calibration technique can be overcome by the proposed FDC-based autocalibration technique, which is described in this section. Figure 10 shows the architecture of the fractional-N PLL with the proposed FDC-based all-digital autocalibration scheme. The calibration circuit comprises a high-speed FDC, the VCO frequency calibration circuit, the loop gain calibration circuit and the auxiliary timing control logic, and calibration voltage generator.
PLL block diagram with the FDC-based autocalibration circuit.
The high-speed FDC enables the fast calibration of VCO frequency and loop gain. The FDC operates in the time period of k·TREF to convert the VCO frequency to a digital value with a conversion resolution of fREF/k whereas the conventional technique in Figure 3 requires N times longer time N·k·TREF for acquiring the same resolution. The detailed block diagram of the FDC is shown in Figure 11(a). The FDC accepts the VCO signal and divides it down to fVCO/4 by using a quadrature-phase predivider, which is composed of a true single phase clock (TSPC) divide-by-2 and a quadrature-phase divide-by-2 as shown in Figure 11(b). After the predivider, the subsequent four counters that is connected to each of the quad-phase output signals count the VCO frequency. As a result, each counter can operate at a reduced speed of fVCO/4 while the overall frequency resolution is not diminished. Finally the sum of the four counter outputs represents the VCO frequency.
(a) Detail block diagram of the FDC. (b) Circuit schematic of quadrature-phase predivider having division ratio of 4.
The frequency error is generated by subtracting the target ftarget digital code from the current fVCO digital code. The resulting sign bit is used as Fast/Slow flag for the binary search process. The absolute value of the frequency error is utilized for the final optimal code selection process to find the final code that has shown the minimum frequency error. The more detailed description of the proposed VCO frequency calibration technique can be found in [14]. After the VCO frequency calibration is completed, the loop gain calibration process begins. It is basically based on the on-chip measurement of KVCO. The FDC extracts two VCO frequencies fVH and fVL at two different tuning voltages VtH and VtL. Then, KVCO is computed digitally. With the computed KVCO, the optimal charge pump gain is generated according to the relation ICP~N/KVCO. After these two autocalibration processes, the normal PLL locking process begins. Proper selection of VtL and VtH is important in the proposed calibration technique. In the chosen range, the charge pump current must be sufficiently constant for accurate loop gain calibration. At the same time, the VCO sub-band tuning curves must be sufficiently overlapping to ensure Vtune to remain in this range after the PLL locking. Considering these aspects, VtL and VtH are chosen to be 0.4 and 0.8 V, respectively, which is 0.4 V off from the 1.2-V supply voltage.
Figure 12(a) shows the calibration procedure by illustrating the waveforms of fVCO, Vtune, and ICP, and Figure 12(b) illustrates how the VCO output frequency wanders on the VCO tuning curve plane during the calibration process. Note that this illustration is only for 4-bit cap bank case. In the first VCO frequency calibration mode, the binary search is performed with Vtune fixed at VDD/2. The final optimal code is fed to VCO at tVFC. In the subsequent loop gain calibration mode, Vtune is sequentially switched from VtL to VtH to extract fVL and fVH, and finally producing KVCO. After tLGC, the optimal charge pump code Icpopt is generated to control the loop bandwidth. After the two-step open-loop calibration is completed, the PLL loop is closed and a normal locking process starts.
(a) Transitions of fVCO, Vtune, and ICP during calibration and locking process. (b) Transition of fVCO during calibration.
In order to verify the proposed autocalibration method, various behavioral simulations were performed for a PLL built in a mixed-signal circuit environment including verilog codes and circuit elements. Figure 13 shows a simulation result of the VCO calibration with a 6-bit cap bank. Due to the widely varying sub-band spacing fstep, the frequency resolution is adjusted by varying k from 1 to 4. Thus, the calibration time varies from 1.05 to 2.03 μs. Figure 14(a) shows a simulated result of Vtune and VCO cap bank code during the full autocalibration process. Vtune is fixed at VDD/2 during the first VCO frequency calibration, and the final code found is 0010100. At the subsequent loop gain calibration, Vtune is switched from 400 to 800 mV. In this time, the FDC extracts two frequencies to calculate KVCO. The simulated loop gain calibration time is 5 μs. Figure 14(b) shows the Vtune waveform during the total locking process at the target frequency of 3900 MHz. All of the above behavioral simulation results prove that the proposed FDC-based autocalibration method works successfully for the VCO frequency and loop gain calibrations.
Behavioral simulation result of the VCO frequency calibration with k-value varying from 1 to 4 according to the required resolution.
Behavioral simulation results for the proposed autocalibration. (a) Vtune and VCO cap bank code during autocalibration. (b) Vtune for the total locking process for 3900 MHz target frequency.
A prototype chip including only the VCO frequency calibration circuit is fabricated in 0.13-μm CMOS technology [14]. The chip micrograph is shown in Figure 15. The die size including the pad frame is 1.33 × 1.58 mm2, and the active area of the VCO calibration circuit is 240 × 400 μm2. It dissipates 15.8 mA from a 1.2 V supply.
Chip micrograph of the ΔΣ fractional-N PLL synthesizer including only the VCO frequency calibration circuit [14].
Figure 16 shows the measured fVCO waveform during the VCO frequency calibration process at 2620 MHz. The measured calibration time is 2.03 μs, and the finally selected optimal code is 20, which show successful operation of the proposed optimal code selection algorithm. The PLL output spectrum at the output frequency of 1492 MHz is shown in Figure 17(a). The reference and fractional spurs are −63 and −68 dBc, respectively. The reference spur appears at the reference frequency of 19.2 MHz. Figure 17(b) shows the measured phase noise, which is −102.1 and −124.1 dBc/Hz at 100 kHz and 1 MHz offset, respectively. The in-band phase noise is −86.5 dBc/Hz at 10 kHz offset. And another prototype PLL chip including the full autocalibration capability has been fabricated and planned to be characterized.
Measured VCO frequency during the VCO frequency calibration.
The linearized coarse tuning characteristics are obtained by proposing a pseudoexponential capacitor bank structure [2] while the conventional cap bank produces a linear capacitance variation. Figure 18(a) shows the measured frequency tuning characteristics of the VCO with the proposed cap bank. Notice that the 64 sub-band tuning curves are evenly distributed. Figure 18(b) gives the plots for the fVCO, KVCO, and fstep over the 64 cap bank codes. For the sake of comparison, the measured results for a test VCO employing a conventional binary-weighted cap bank structure are shown together. It can be observed that the proposed cap bank structure substantially reduce the KVCO variation from 38–180 (473%) to 34–90 (264%) MHz/V, and the fstep variation from 5.6–31 (553%) to 6–18 (300%) MHz/code. The results prove that the proposed pseudoexponential cap bank structure effectively linearizes the coarse tuning characteristics and reduces the KVCO and fstep variations. It is found to be very instrumental for the autocalibration of a PLL having a wide-tuning range LC VCO.
Measured VCO tuning characteristics. (a) Frequency tuning curves. (b) VCO output frequency, KVCO, and fstep over the 64 tuning curves.
4. Conclusion
Design considerations for the autocalibration of VCO frequency and loop gain are discussed for wide-band ΔΣ fractional-N PLL synthesizers. To overcome the limitations of the conventional techniques, the FDC-based all-digital autocalibration circuit is proposed. The FDC-based method remarkably shortens calibration time and improves calibration accuracy, and was found to be highly suitable for a wide band ΔΣ fractional-N PLL.
Acknowledgment
This paper has been supported by the ITRC Program (NIPA-2011-C1090-1111-0006).
KralA.BehbahaniF.AbidiA. A.RF-CMOS oscillators with switched tuningProceedings of the IEEE Custom Integrated Circuits ConferenceMay 19985555582-s2.0-0031629744KimJ.ShinJ.KimS.ShinH.A wide-band CMOS LC VCO with linearized coarse tuning characteristics20085553994032-s2.0-4484913072710.1109/TCSII.2007.914896WilsonW. B.MoonU. K.LakshmikumarK. R.DaiL.A CMOS self-calibrating frequency synthesizer20003510143714442-s2.0-003429374210.1109/4.871320LinT. H.KaiserW. J.A 900
MHz 2.5
mA CMOS frequency synthesizer with an automatic SC tuning loop20013634244312-s2.0-003527450510.1109/4.910481AktasA.IsmailM.CMOS PLL calibration techniques20042056112-s2.0-724424053510.1109/MCD.2004.1343243LeeH. R.HwangM. S.LeeB. J.KimY. D.OhD.KimJ.LeeS. H.JeongD. K.KimW.A 1.2
V-Only 900
mW 10
Gb ethernet transceiver and XAUI interface with robust VCO tuning technique20054011214821572-s2.0-2784459166910.1109/JSSC.2005.857360LeeH. I.ChoJ. K.LeeK. S.HwangI. C.AhnT. W.NahK. S.ParkB. H.A Σ-Δ fractional-N frequency synthesizer using a wide-band integrated VCO and a fast AFC technique for GSM/GPRS/WCDMA applications2004397116411692-s2.0-304282214710.1109/JSSC.2004.829938MarutaniM.AnbutsuH.KondoM.ShiraiN.YamazakiH.WatanabeY.An 18mW 90 to 770MHz synthesizer with agile auto-tuning for digital TV-tunersProceedings of the IEEE International Solid-State Circuits Conference, (ISSCC '06)February 20061911922-s2.0-39749096822LinT. H.LaiY. J.An agile VCO frequency calibration technique for a 10
GHz CMOS PLL20074223403492-s2.0-3384773029510.1109/JSSC.2006.889360LeeJ.KimK.LeeJ.JangT.ChoS.A 480-MHz to 1-GHz sub-picosecond clock generator with a fast and accurate automatic frequency calibration in 0.13-μm CMOSProceedings of the IEEE Asian Solid-State Circuits Conference, (A-SSCC '07)November 200767702-s2.0-5134912767210.1109/ASSCC.2007.4425733AkamineY.KawabeM.HoriK.OkazakiT.KasaharaM.TanakaS.ΔΣ PLL transmitter with a loop-bandwidth calibration system20084324975062-s2.0-3884911182710.1109/JSSC.2007.914325ShananH.RetzG.MulvaneyK.QuinlanP.A 2.4
GHz 2
Mb/s versatile PLL-based transmitter using digital pre-emphasis and auto calibration in 0.18μm CMOS for WPANProceedings of the IEEE International Solid-State Circuits Conference (ISSCC '09)February 20092-s2.0-7034926972610.1109/ISSCC.2009.4977487WuT.HanumoluP. K.MayaramK.MoonU. K.Method for a constant loop bandwidth in LC-VCO PLL frequency synthesizers20094424274352-s2.0-5934911541710.1109/JSSC.2008.20107924768902ShinJ.ShinH.A fast and high-precision VCO frequency calibration technique for wideband ΔΣ fractional-N frequency synthesizers2010577157315822-s2.0-7795486483610.1109/TCSI.2009.20360575371871RheeW.AinspanH.FriedmanD. J.RasmusT.GarvinS.CranfordC.A uniform bandwidth PLL using a continuously tunable single-input dual-path LC VCO for 5 Gb/s PCI express Gen 2 applicationProceedings of the IEEE Asian Solid-State Circuits Conference, (A-SSCC '07)November 200763662-s2.0-5134909467510.1109/ASSCC.2007.4425732HauspieD.ParkE. C.CraninckxJ.Wideband VCO with simultaneous switching of frequency band, active core, and varactor size2007427147214802-s2.0-3434720832710.1109/JSSC.2007.899105VäänänenP.MikkolaN.HeliöP.VCO design with on-chip calibration system20065310215721662-s2.0-3375042446710.1109/TCSI.2006.883844LuL.ChenJ.YuanL.MinH.TangZ.An 18-mW 1.1752-GHz frequency synthesizer with constant bandwidth for DVB-T tuners20095779289372-s2.0-6474910222510.1109/TMTT.2009.20144494799232