California Eastern Laboratories APPLICATION NOTE AN1026 1/f Noise Characteristics Influencing Phase Noise Abstract In applications such as VCO’s where 1/f noise is one of the major contributors to phase noise, the device with the best 1/f noise will yield the best phase noise. To better understand how 1/f noise changes with bias and from device to device, the 1/f noise of two different bipolar devices were measured and compared. Nonlinear models were developed and the predicted and measured 1/f noise spectrums compared. In addition, how manufacturers quantify 1/f for simulation purposes is evaluated to provide the designer who is new to noise simulation with better insight into what the 1/f coefficient and exponent mean, and how the values effect the 1/f noise simulations. I. Introduction The phase noise of an oscillator, quantified by its short term frequency stability, determines a system’s ability to separate adjacent channels. With the significant increase in portable, wireless traffic, better frequency stability is required. Phase noise becomes an important consideration. Choosing the appropriate topology for a low phase noise source is important, as is the choice of resonator and coupling network [1]. Many articles exist on how to design an oscillator for low phase noise by selecting the appropriate topology, resonator and coupling network[2]. This article focuses on the important characteristics for device evaluation and selection. 1 Hz Fosc Figure 1. Single sideband phase noise Oscillator phase noise has two components: phase noise resulting from direct upconversion of white noise and 1/f noise, and phase noise resulting from the changing phase of the noise sources modulating the oscillation frequency, ∆ freq. The composite oscillator phase noise spectrum consists of the white noise and 1/f noise directly upconverted around the carrier, and the phase modulated noise spectrum. The frequency noise in the frequency domain is converted to phase noise in the frequency domain by multiplying by 1/f2. The white noise spectra, f0, then contributes 1/f2 and the 1/f noise spectra contributes 1/f3, see Table 1. Noise Source Phase Noise Spectrum Upconverted II. Oscillator Phase Noise Phase perturbations due to random noise in the oscillator result in a random shift of the oscillator frequency. These random frequency variations are caused by thermal noise, shot noise, and flicker noise. Thermal noise is a function of the temperature, bandwidth and resistance, shot noise is a function of the DC bias current, and flicker noise is a function of the active device characteristics. Single sideband phase noise is defined as the noise power in a 1-Hz bandwidth at some offset frequency from the carrier, Fosc, and has the units of dBc/Hz, Figure 1. ∆ freq white noise white noise 1/f2 1/f noise 1/f noise 1/f3 Table 1. Noise sources and their phase noise spectrum AN1026 SSB Phase Noise (dBc/Hz) For a low Q oscillator where the half-bandwidth frequency, fo/2Q, is greater than the flicker corner frequency, fc, the oscillator phase noise characteristics will be determined by the white noise, 1/f2, and the flicker noise, 1/f3, Figure 2. -60 -80 f -3flicker FM S(f)df = (KF)IBAFdf / Fc -100 where: KF AF IB Fc f -2 -120 white FM -140 -160 -10 fc 1k 100 10k fo/2Q 100k Baseband offset Frequency (Hz) Figure 2. Baseband noise spectrum for fc<fo/2Q, Low Q. SSB Phase Noise (dBc/Hz) For a high Q oscillator where the halfbandwith frequency, fo/2Q, is less than the flicker corner frequency, fc, the oscillator phase noise characteristics will be determined by the frequency modulated flicker noise, 1/f-3, and the upconverted 1/f noise, Figure 3. -60 (1) = flicker noise constant = flicker noise exponent = DC base current = flicker noise corner frequency A. Noise Test Setup The test setup to measure 1/f noise is shown in Figure 4. The 50 ohm terminations insure amplifier stability at higher frequencies. The entire test setup is contained in an EMI shielded box. The supply voltages, VBB and VCC are provided by batteries to eliminate 60 Hz noise. Vcc LPF VBB -80 -100 Flicker noise in BJTs is also known as 1/f noise because of the 1/f slope characteristics of the noise spectra. This noise is caused mainly by traps associated with contamination and crystal defects in the emitter-base depletion layer. These traps capture and release carriers in a random fashion. The time constants associated with the process produce a noise signal at low frequencies. The flicker noise spectral density is given by: FET Test Setup CPU/ Printer LPF f -3flicker FM 50 f -1 -120 50 -140 -160 fo/2Q -10 100 1k 10k Baseband offset Frequency (Hz) 100k Figure 3. Baseband noise spectrum for fc>fo/2Q, High Q. III. 1/f Measurements 1/f measurements are performed to obtain the flicker noise frequency spectra and determine the flicker corner frequency. From this spectra, the simulation coefficient KF and the simulation exponent AF necessary for modeling the 1/f noise spectra can be determined. Figure 4. 1/f Noise Test Setup The measured 1/f noise spectrum for the NE68819 at Vce = 3V, Ic=12mA is shown in Figure 5. The measured flicker corner frequency, Fmeas = 5.6kHz, is determined by noting the intersection of the 1/f noise spectrum and the white noise spectrum. This intersection is where the measured flicker noise power and the white noise power are equal. AN1026 Calculated AF and KF values resulting from 1/f measurements for the NE68519 and the NE68819 are provided in Table 2. -110 NE68519 -120 Ic (mA) -130 -140 -150 -160 -170 110 1K 5.6 KHz 10K 100K Bv(f) [dBv/Hz] vs. f[Hz] NE68819 VCE=1V AF KF VCE=3V AF KF VCE=1V AF KF VCE=3V AF KF 5 1.56 454 1.35 80 1.9 5 1.18 10 10 1.48 231 1.19 18 1.39 72 1.22 15 20 1.89 7.6K 1.12 13 1.19 14 1.23 16 30 1.6 774 1.25 30 1.16 10 1.26 26 Table 2. AF and KF for the NE68519 and the NE68819 (KF values are times 10-15) C. Bias Dependency Figure 5. 1/f measurement for the NE68819 at Vce=3V, Ic=12 mA To determine the intrinsic base flicker noise corner, Fbn, requires solving the following equation [3]: Fbn = Fmeas [1 + 1/β + 2VthGin/IB] (2) where: Fbn = intrinsic base flicker noise corner Fmeas = measured flicker corner β = collector-base current gain Vth = thermal voltage = kT/q Gin = external input conductance IB = DC base biasing current Next we examine the bias dependency of the flicker corner frequency for the same two NEC devices. The base voltage was varied to obtain the collector currents shown in Table 3. The flicker corner frequency, Fbn, was determined at VCE=1V and VCE=3V as shown in Table 3 and Figure 6. Fbn (KHz) Ic (mA) 5 The equation for the intrinsic base flicker corner modifies the measured flicker corner to account for the input conductance, base current, and DC current gain of the device. The formula for Fbn is valid providing the measured output noise characteristics are dominated by the base flicker and base shot noise sources. The design of the test setup and selection of biasing circuitry is chosen to insure these conditions. NE68519 Vce=1V Vce=3V 5.8 7.3 NE68819 Vce=1V Vce=3V 5.6 5.1 10 8.8 9.9 5.9 5.8 20 12.4 11.2 7.7 6.7 30 18.6 11.9 8.3 7.3 Table 3. Fbn versus bias - NE68519, NE68819 20 Measuring the flicker corner frequency at two different base currents provides two different flicker corner values. Using equation (3) twice, once for each IB and Fbn value provides two equations and two unknowns, resulting in a solution for AF and KF. Fbn = (KF)IB(AF-1) / 2q (3) Fbn (KHz) 15 B. Calculating AF and KF 10 5 0 0 5 10 15 20 Ic (mA) 25 30 NE68519 NE68819 The resulting expressions for AF and KF are: AF = log(F1) - log(F2) + Z Z Z = log (IB1) - log (IB2) (4) KF = 10 (log (Fbn) + log (IB) -18.49 - (AF)log (IB)) (5) Figure 6a. Fbn at VCE=1V for the NE68519 and the NE68819 35 Fbn (KHz) AN1026 20 -110 15 -120 10 -130 5 0 -140 0 5 10 15 20 Ic (mA) 25 30 35 -150 -160 NE68519 NE68819 -170 Figure 6b. Fbn at VCE=3V for the NE68519 and the NE68819 100 5 9.5K 1K 10K 5V (f) [dBv/Hz] vs. f (Hz) 100K The flicker corner frequencies for the two devices are within 3% at the lower bias of Vce=1V and Ic=5mA; Fbn = 5.8kHz for the NE68519 and Fbn=5.6kHz for the NE68819. At higher biases, the NE68819 has lower 1/f noise. Figure 7b. 1/f measurement with Vce constant, Vce=3V, and Ic=5mA aand 10mA for the NE68519 As the bias voltage increases from 1V to 3V with a constant collector current Ic=20mA, the amplitude of the 1/f slope is unchanged but the amplitude of the white noise floor increases due to the increase in thermal noise voltage as shown in Figure 7a. The higher noise floor results in a lower 1/f corner frequency of 11.8kHz at Vce=3V. At Vce=1V the 1/f corner frequency is 16kHz. A nonlinear model was developed for the NE68519 and the NE68819. The simulation schematic and the variable values for the NE68819 are shown in Figures 8a and 8b. The simulator used is the LIBRATM harmonic balance simulator by HP/EESOF [4]. The test bench is shown in Figure 8c. Note that AF=1.26 and KF=2.6e-14 to provide flicker noise simulation. Simulated 1/f noise for the NE68819 at Vce=3V, Ic=10mA is shown in Figure 9. IV. 1/f Noise Simulation A. NE68819 simulated versus measured -110 -120 -130 -140 3V 1V -150 Measured 1/f noise spectra at Vce=3V, Ic=12mA, Figure 5, was compared to simulated 1/f noise at Vce=3V, Ic=10mA, Figure 9. Although the collector currents were 20% different, the 1/f corner frequencies - measured vs model - were within 10% of each other. The white noise floor was -149dB (measured) compared to -151dB (model) resulting in a 1.32% difference. The 1MHz noise amplitude was -131dB (measured compared to -132dB (modeled), a difference of only 0.76%. -160 B. Modeling 1/f noise with AF and KF -170 100 1K 10K 9V(f) [dBv/Hz] vs. f[Hz] 100K Figure 7a. 1/f measurement with IC constant, Ic=20mA and Vce=1V and 3V for the NE68519 As the bias current increases from 5mA to 10mA and the bias voltage stays constant at Vce=3V, the 1/f spectrum amplitude increases proportionally with increases in base current, and the shot noise component of the white noise floor increases proportionally with increasing base current, Figure 7b. Most commercial simulators set AF and KF to default values of 1.0 and 0 respectively. Inspection of equation (1) shows that if KF = 0, the flicker noise spectral density will also be 0, resulting in the presence of white noise only. Examining figure 10c with KF=1e-13 shows the 1/f spectra approaching the white noise spectra. Figures 10a and 10b model the 1/f spectra changes as KF and AF are varied. Note that the Yintercept of the 1/f spectra increases proportionally to KF, as expected from equation (1). The Y-intercept of the 1/f spectra decreases more rapidly with increases in AF. Equation (2) verifies this to be the case with the 1/f spectral density decreasing exponentially with increases in AF, since IB is less than unity. AN1026 CAP Co C = 1000000000 AMMETER AMM1 RES RL R = 2000 TEMP = TEMP2 IND Lc L = Lc CAP Ccb C = Ccb PORT P1 port = 1 CAP Ci_g C= 1000000000 RES AMMETER Rbb AMM2 R = rbb TEMP = TEMP2 CAP Ccc C = Ccc IND Lb L = Lb BJT BJT1 AREA = 1 MODEL = NE688 chip model MODE = nonlinear + DCVS SRC1 _ DC = 5.48 PORT P2 port = 2 CAP Cc1 C = 1000000000 IND La L = 1000000000 + DCVS SRCc - DC = 23 IND Le L = Le BJTM NE688 chip model IS = is BR = br BF = bf NR = nr NF = nf VAR = var VAF = vaf IKR = ikr IKF = ikf ISC = isc ISE = ise NC = nc NE = ne RB = rb IRB = irb RBM = rbm RE = re RC = rc CJE = cje VJE = vje MJE = mje TF = tf XTF = xtf VTF = vtf ITF = itf PTF = 0 CJC = cjc VJC = vjc MJC = mjc XCJC = xcjc TR = tr CJS = 0 VJS = 0.75 MJ5 = 0 XTB = 0 EG = 1.11 XT = 3 KF = kf AF = af FC = fc TYPE = type NK = 0.50 TNOM = 27 FFE = 1 KB = 0 AB = 1 FB = 1 ISS = 0 NS = 1 Figure 8a. NE68819 Simulation Schematic DATA UNITS UNITS_DEFAULT FREQ=GHz RES=0hm COND=S IND=nH CAP=pF LNG=mil TIME=psec ANG=deg POWER=dBm VOLT=V CUR=mA DIST=mi TEMP TEMP_DEFAULT TEMP=27 DATA DATA DATA RREF RREF_DEFAULT R=50 TAND TAND_DEFAULT TAND=0 SIGMA SIGMA_DEFAULT SIGMA=0 tf=1.10e-11 xtf=0.36 vtf=0.65 itf=0.61 ptf=50 cjc=5.49e-13 vjc=0.65 mjc=0.48 xcjc=0.56 tr=3.20e-11 fc=0.75 kf=2.60e-14 cf=1.26 Ccb=0.24 Cce=0.27 Lb=0.72 Lc=0.51 Le=0.19 rbb=50000 DATA TEMP TEMP2 TEMP=0 Var Eqn VAR _VAR is=3.80e-16 bf=135.70 nf=1.00 vaf=28 ikf=0.60 ise=3.80e–15 ne=1.49 br=12.28 nr=1.10 var=3.50 ikr=0.06 isc=3.50e-16 nc=1.62 rb=6.14 irb=1.00e-03 rbm=3.50 re=0.40 rc=4.20 cje=7.96e-13 vje=0.71 mje=0.38 NE688_chip_model PARAMETERS AND BONDING PARASITICS DATE: 10/2/96 PROJECT NAME: 68800 LIBRA DEFAULT FILE: 68800_def Figure 8b. NE68819 nonlinear model parameter values 68800_noise_sch X1 FREQUENCY FPLAN value = ESWEEP 1.00e-07 1.00e-04 50 VNOISEQ VNOISEQ1 PORT = 2 OUTPUT EQUATION OUT EQN OUTEQN v_noise_out = 20* log(VNOISEQ1) Figure 8c. NE68819 test bench DATA PERM PERM_DEFAULT MUR=1 TANM=0 AN1026 68800_mod_noise10_tb v_noise_out 68800_noise OUT EQN Re -100.0 -105.0 -110.0 -115.0 -120.0 -125.0 -132 dB -130.0 -131 dB -135.0 -140.0 -145.0 -150.0 -155.0 -160.0 -165.0 -170.0 0.1 -100.0 -110.0 Measured KF = 20e-13 -120.0 Model KF = 10e-13 -130.0 KF = 5e-13 -151 dB -149 dB KF = 2e-13 KF = 1e-13 -140.0 -150.0 5.8 KHz 6.3 KHz Frequency KHz NE68800 noise voltage - effect of varying AF & KF VCE = 3 V IC = 10 mA -160.0 100.0 Model Measured -170.0 0.1 Figure 9. 1/F noise simulation for the NE68819 at Vce= 3V, Ic=10 mA Frequency KHz 100.0 NE68800 noise voltage - effect of varying AF & KF AF = 1.6, Vce = 3V, KF = 1e-13 to 2e-12 Ic = 10mA Figure 10c. Effect of varying KF from 1e-13 to 20-e-13 with AF=1.6 CONCLUSIONS 1. -100.0 AF = 1 -110.0 2. -120.0 -130.0 3. -140.0 -150.0 4. AF = 2 -160.0 -170.0 0.1 100 Frequency KHz NE68800 noise voltage - effect of varying AF & KF AF = 1 to 2, KF = 5e-13 Vce = 3V Ic = 10ma Figure 10a. Effect of varying AF from 1 to 2 with KF=5e-13 -100.0 -110.0 REFERENCES [1] Roger Muat, “Choosing devices for quiet oscillators”, Microwave & RF, August 1984, pp. 166-170. [2] Grant Moulton, “Dig for the roots of oscillator noise”, Microwave & RF, pp. 65-69, April 1986. [3] Julio Costa et al, “Extracting 1/f Noise Coefficients for BJT’s,” IEEE Transactions on Electron Devices, Vol. 41, No.11, pp. 1992-1999, Nov. 1994 [4] “Simulating Noise in Nonlinear Circuits Using the HP Microwave and RF Design Systems,” HP Product note 851804, c 1993. KF = 20e-13 KF = 10e-13 KF = 5e-13 -120.0 KF = 2e-13 -130.0 -140.0 KF = 1e-13 -150.0 California Eastern Laboratories -160.0 -170.0 0.1 5. Using the test setup and measurement techniques described, 1/f noise can be measured. Using an accurate nonlinear model, 1/f noise simulations agree closely to measured data. As Vce increases, the flicker corner increases as the white noise increases, but the magnitude of the 1/f noise is constant. As base current increases, the flicker corner frequency increases with the magnitude of the 1/f noise and the increased shot noise current. 1/f noise increases linearly as KF increases and decreases with increases in AF for IB less than 1.0 amp, as is expected from equation (1). Frequency KHz 100.0 NE68800 noise voltage - effect of varying AF & KF AF = 1, KF = 1e-13 to 2e-12 Ic = 10mA Figure 10b. Effect of varying KF from 1e-13 to 20-e-13 with AF=1 Exclusive Agents for NEC RF, Microwave and Optoelectronic semiconductor products in the U.S. and Canada 4590 Patrick Henry Drive, Santa Clara, CA 95054-1817 Telephone 408-988-3500 • FAX 408-988-0279 •Telex 34/6393 Internet: http:/WWW.CEL.COM Information and data presented here is subject to change without notice. California Eastern Laboratories assumes no responsibility for the use of any circuits described herein and makes no representations or warranties, expressed or implied, that such circuits are free from patent infingement. © California Eastern Laboratories 02/04/2003