Berkeley Mixer Noise Prof. Ali M. Niknejad U.C. Berkeley c 2014 by Ali M. Niknejad Copyright Niknejad Advanced IC’s for Comm Noise in Mixers Prof. Ali M Niknejad and Dr. Osama Shana’a Niknejad Advanced IC’s for Comm Noise in an Ideal Mixers Consider the simplest ideal multiplying mixer: RF Noise IF LO IF RF LO IM What’s the noise figure for the conversion process? Input noise power due to source is kTB where B is the bandwidth of the input signal Input signal has power Ps at either the lower or upper sideband (or both for direct conversion) SNRi = Ps kTB Niknejad Advanced IC’s for Comm Noise in Ideal Mixers At the IF frequency, we have the down-converted signal G · Ps and down-converted noise from two sidebands, LO − IF and LO + IF SNRo = (G 0 G · Ps + G 00 ) kTB For ideal mixer, G = G 0 = G 00 F = Ps 2kTB SNRi = =2 SNRo kTB Ps NF = 3dB For a real mixer, noise from multiple sidebands can fold into IF frequency and degrade NF Niknejad Advanced IC’s for Comm SSB vs. DSB NF definition: image noise LO Ni+Gmix Ni S/Ni image noise+Gmix IF IF As we have seen, because of the image problem, a receive mixer down converts both desired and the image bands to IF frequency. This means folding the noise at the image frequency on top of the desired band at IF. Therefore, the total noise at IF is as follows: The noise at desired RF band down converted to IF The noise at image RF band down converted to IF The noise added by the mixer noisy circuit itself. Niknejad Advanced IC’s for Comm Single-Side Band (SSB) NF The single-side band NF definition assumes that there is no signal at the image frequency except the source noise. This definition is useful in non-zero IF architectures, where the image signal is suppressed by an image filter before reaching the mixer. The NF is the degradation of S/N at mixer output. Therefore, one can write: Sout = Sd Gmix,d Nout = Nd Gmix,d +NI 0 m Gmix,I 0 m +Nmix,d Gmix,d +Nmix,im Gmix,I 0 m where Sd is the desired signal, Nd is the noise in the desired band, NI 0 m is the noise in the image band, Gmix,d and Gmix,im is the mixer gain in the desired and image frequencies, respectively. Nmix,d and Nmix,I 0 m are the desired and image band noise due to the mixer circuit itself referred to its input. Niknejad Advanced IC’s for Comm Single-Side Band (SSB) NF (cont) To simplify the analysis, we will assume that Nd = NI 0 m , Nmix,d = Nmix,im , and Gmix,d = Gmix,I 0 m , so that we can write ! Sd Sd 1 Sout = = Nout 2Nd + 2Nmix Nd 2 + 2NNmix d FSSB = 2 + 2Nmix Nd As seen from the SSB noise figure equation, if the mixer is noiseless (Nmix = 0), the mixer SSB NF is 3dB because of the image noise folding. It is important to know that this definition is the one used by microwave mixer designers for years. It is also the definition used in the SpectreRF simulator. Niknejad Advanced IC’s for Comm IEEE Noise Definition The IEEE has a slightly different definition for SSB NF. It argues that the mixer should not be “penalized” by the image source noise folding. The only image noise folding that is allowed to count towards calculating the mixer SSB NF is that which is due to the mixer circuitry itself. The input image noise should not be counted. Therefore, the IEEE SSB NF assumes there is a sharp bandpass filter that passes the desired band with the source noise and knocks down the image noise to the negligible level. As a result, one can write Sout = Sd Gmix,d Nout = Nd Gmix,d + Nmix,d Gmix,d + Nmix,im Gmix,im To simplify analysis, we will assume Nd = Nim and Gmix,d = Gmix,im , therefore, Nout = Nd Gmix,d + 2Nmix Gmix,d Niknejad Advanced IC’s for Comm IEEE Noise Def (cont) Image noise filter image noise LO Noiseless mixer Nd+Gmix Sd/Nd Nd IF IF # " ⇒ Sout Nout = ⇒ FSSB Sd Nd +2Nmix IEEE =1+ = Sd Nd 1 1+ 2Nmix Nd 2Nmix Nd As seen from the IEEE SSB equation, if the mixer is noiseless, the mixer SSB NF is actually 0dB, the spirit behind the new definition. Niknejad Advanced IC’s for Comm Double-Side Band (DSB) NF: LO Noiseless mixer image Nd+Gmix_d image noise Nd Sd/Nd Nim+Gmix_im IF IF The double-side band NF definition assumes that the image band contains both noise and an image signal identical to the desired band signal. This definition is useful in direct-conversion receiver where the image is the signal itself. Therefore, one can write: Sout = Sd Gmix Nout = Nd Gmix d +Nim Gmix d + Sim Gmix im im +Nmix d Gmix d +Nmix im Gmix im where Sim is the image signal. To simplify analysis, we will assumeSd = Sim , Nd = Nim , Nmix,d = Nmix,im , and Gmix,d = Gmix,im . Therefore, Nout = 2Nd Gmix,d + 2Nmix Gmix,d Niknejad Advanced IC’s for Comm DSB NF (cont) " ⇒ Sout Nout = 2Sd 2Nd +2Nmix ⇒ FDSB = 1 + = Sd Nd # 1 1+ Nmix Nd Nmix Nd It can be seen that the difference between the SSB NF and DSB NF is exactly 3dB. However, with the SSB IEEE, the difference is not exactly 3dB. In fact the difference between the SSB IEEE and the SSB NF approaches 3dB as the mixer NF is very high. The SSB IEEE noise factor can be related to that of the DSB as: FSSB IEEE Niknejad = 2FDSB − 1 Advanced IC’s for Comm Noise in Active Mixers Prof. Ali M Niknejad Based on the work of Dr. Manolis Terrovitis, “Analysis and Design of Current-Commutating CMOS Mixers,” Ph.D. Dissertation, Berkely, CA 2001. Niknejad Advanced IC’s for Comm Recap: CMOS Mixer Operation I1 I2 M2 Io1 = I1 − I2 = F (VLO (t) , IB + is ) ∂F ≈ F (VLO (t) , IB ) + (VLO (t) , IB ) · is + · ∂IB = p0 (t) + p1 (t) · is + · · · M3 LO RF IB + i s M1 Periodic function: p1 (t) = gm1 (t) − gm2 (t) gm1 (t) + gm2 (t) p1 (t) = −p1 (t + TLO /2) with good matching, which means it has only odd-order frequency components. Fourier Series expansion P1,2k ≡ 0 Niknejad Advanced IC’s for Comm Mixer Noise AC signal at output is given by Io1 = p1 (t) · is Let X (f ) be the spectrum of is , Y (f ) be the spectrum of Io1 ∞ X Y (f ) = n=−∞ P1,n X (f − nfLO ) If we fix the output frequency at fIF , then we see energy from multiple bands folding into output spectrum n e o is no −3fo −2fo fold is in g o ld ef in g −fs −fo i s e f o l d i n g f ro m no 3LO n o i s e fo l di n gf rom 2L O from -3LO fro m - 2 LO −fif Niknejad fif fo fs Advanced IC’s for Comm 2fo 3fo Mixer Noise (cont) Y (fIF ) = ∞ X n=−∞ p1,n X (fIF − nfLO ) n = +1 n = −1 n = +2 fLO − fIF fLO + fIF 2fLO − fIF The coefficient p1,k then represents the conversion gain from frequency (k × fLO ± fIF ) to fIF . If we assume hard switching, then p1 (t) is a square wave and the coefficients are 2 p1,2k−1 = π (2k − 1) Niknejad Advanced IC’s for Comm Noise from Transconductance Stage A cyclostationary process is a random process whose statistics are periodic functions of time. The PSD of a cyclostationary process is given by S(f , t). If we measure a cyclostationary process over a bandwidth < 1/period, we observe S (f , t), a stationary process. Here we assume that the noise of the transconductor is stationary. The output noise, though, is cyclostationary since it’s multiplied by a periodic function. n3 (t) WSS, PSD sn3 (f ) Yn3 = n3 (t) · p1 (t) Niknejad Advanced IC’s for Comm PSD Calculation Consider a white noise source n(t) multiplied by a periodic deterministic function p(t): y (t) = n (t) · p (t) Ry (t1 , t2 ) = E [y (t1 ) y (t2 )] = p (t1 ) p (t2 ) n (t1 ) n (t2 ) = p (t1 ) p (t2 ) δ (t1 − t2 ) · N Ry (τ ; t) = p(t)2 δ (τ ) · N Sy (f ; t) = Np 2 (t) Z Sy (f ; t) = N p 2 (t) dt T Niknejad Advanced IC’s for Comm PSD (cont) If Sn3 (t) = Nn3 (white), then α= 1 TLO Sn3 (f ) = Nn3 · α Z ∞ X 2 p1 (t) dt = |p1,n |2 TLO Niknejad n=−∞ Advanced IC’s for Comm Square LO Waveform Noise For a square wave p1 (t), α = 1 First sidebands (fLO ± fIF ) accounts for 2 2 ≈ 81% 2× π (3fLO ± fIF ) accounts for 2× 2 3π 2 ≈ 7% The remaining harmonics account for the rest ∼ 10% Niknejad Advanced IC’s for Comm Nearly Square LO Waveform vo vx -vx -vo p1 (t ) +1 Assume p1 (t) is a straight line during period ∆ when both devices on. Then 4 α = 1 − ∆fLO 3 −1 vx π∆fLO = sin vo Niknejad Advanced IC’s for Comm Single Balanced Mixer Noise Given the results so far, we can write the thermal noise due to the transconductance stage as: γ Sn3 (f ) = 4kT Rs + rg 3 + · gm3 2 ·α |{z} | {z } gm3 | {z } 2 Noise due to M3 G for M3 with de(transconductor) Input referred noise voltage m3 generation For a Gilbert Cell, we double the noise contribution due to the presence of two devices with uncorrelated noise: 2γ Sn3 (f ) = 4kT Rs + 2rgs + Gm3 2 · α gm3 Niknejad Advanced IC’s for Comm Thermal Noise of Switching Pair I1 High RF M2 M3 M1 Low When we switch hard, the noise is due to M3 only (transconductor) (neglecting capacitance and output impedance of M3) When we switch hard, the noise is due to M3 only (transconductor) (neglecting capacitance and output impedance of M3) Hard switching is good for low noise output. When both M1 and M2 on, during the switching period ∆, then the noise PSD at output is !2 !2 1 gm1 1 gm2 SI1 = 4kT γ + gm1 m2 gm1 1 + gm2 gm2 1 + ggm1 = 4kT γ Niknejad gm1 · gm2 gm1 + gm2 Advanced IC’s for Comm Thermal Noise of Switching Pair (2) I3 = I1 + I2 Io1,n = 2I1,n ⇒ 4X power Total transconductance of diff pair for LO port to differential output current: Sn1,2 (f , t) = 16kT γ G (t) = 2gm1 gm2 gm1 + gm2 gm1 gm2 = 8kT γG (t) gm1 + gm2 1 Sn1,2 (f ) = Sn1,2 (f , t) = 8kT γ TLO TLO Z G (t) dt 0 = 8kT γG Niknejad Advanced IC’s for Comm Thermal Noise of Switching Pair (3) vLO = vo sin (2πfLO t) 1 G= πvo Zvx G (vLO ) r 1− −vx 1 ≈ πvo Zvx −vx 1 vLO vo 2 dvLO dIo1 2IB · dvLO = dvLO πvo 16kT γ Sn1,2 (f ) ≈ π Niknejad IB vo Advanced IC’s for Comm Noise from LO Port Consider the noise that originates from the LO port, “outside” of the mixer: YnLO = G (t) · nLO (t) SnLO (t) = NLO ∞ X −∞ |Gn |2 = G 2 NLO For example, if we assume the mixer port is driven with an ideal voltage source, then only the gate resistances contribute to the noise: Sn,LO (f ) = 4kT (4rg 1 ) G 2 LO Niknejad Advanced IC’s for Comm Flicker Noise LO DC DC → LO in up-conversion from fLO and all odd harmonics Flicker noise of switching transferred to output by multiplication by G (t) G (t) has period TLO /2, and Fourier series contains only even-order harmonics of the LO. Flicker noise from the switching pair will appear at output at DC but not at fLO . Niknejad Advanced IC’s for Comm Mixer Noise Figure Neglecting Flicker Noise: (NF )SSB α (γ3 + rg 3 gm3 )gm3 α + 2γ1 G + (RLO + 2rg 1 )G 2 + = 2+ 2 R c c 2 gm3 s And for the Gilbert cell is (NF )SSB = α 2(γ3 + rg 3 gm3 )gm3 α + 4γ1 G + (4rg 1 )G 2 + + 2 R c2 c 2 gm3 s Niknejad Advanced IC’s for Comm 1 RL , 1 RL